| 1 | /* |
| 2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
| 3 | * Copyright (c) 2013 Red Hat, Inc. |
| 4 | * All Rights Reserved. |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU General Public License as |
| 8 | * published by the Free Software Foundation. |
| 9 | * |
| 10 | * This program is distributed in the hope that it would be useful, |
| 11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 13 | * GNU General Public License for more details. |
| 14 | * |
| 15 | * You should have received a copy of the GNU General Public License |
| 16 | * along with this program; if not, write the Free Software Foundation, |
| 17 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA |
| 18 | */ |
| 19 | #include "xfs.h" |
| 20 | #include "xfs_fs.h" |
| 21 | #include "xfs_shared.h" |
| 22 | #include "xfs_format.h" |
| 23 | #include "xfs_log_format.h" |
| 24 | #include "xfs_trans_resv.h" |
| 25 | #include "xfs_bit.h" |
| 26 | #include "xfs_sb.h" |
| 27 | #include "xfs_ag.h" |
| 28 | #include "xfs_mount.h" |
| 29 | #include "xfs_da_format.h" |
| 30 | #include "xfs_da_btree.h" |
| 31 | #include "xfs_dir2.h" |
| 32 | #include "xfs_dir2_priv.h" |
| 33 | #include "xfs_inode.h" |
| 34 | #include "xfs_trans.h" |
| 35 | #include "xfs_inode_item.h" |
| 36 | #include "xfs_alloc.h" |
| 37 | #include "xfs_bmap.h" |
| 38 | #include "xfs_attr.h" |
| 39 | #include "xfs_attr_leaf.h" |
| 40 | #include "xfs_error.h" |
| 41 | #include "xfs_trace.h" |
| 42 | #include "xfs_cksum.h" |
| 43 | #include "xfs_buf_item.h" |
| 44 | |
| 45 | /* |
| 46 | * xfs_da_btree.c |
| 47 | * |
| 48 | * Routines to implement directories as Btrees of hashed names. |
| 49 | */ |
| 50 | |
| 51 | /*======================================================================== |
| 52 | * Function prototypes for the kernel. |
| 53 | *========================================================================*/ |
| 54 | |
| 55 | /* |
| 56 | * Routines used for growing the Btree. |
| 57 | */ |
| 58 | STATIC int xfs_da3_root_split(xfs_da_state_t *state, |
| 59 | xfs_da_state_blk_t *existing_root, |
| 60 | xfs_da_state_blk_t *new_child); |
| 61 | STATIC int xfs_da3_node_split(xfs_da_state_t *state, |
| 62 | xfs_da_state_blk_t *existing_blk, |
| 63 | xfs_da_state_blk_t *split_blk, |
| 64 | xfs_da_state_blk_t *blk_to_add, |
| 65 | int treelevel, |
| 66 | int *result); |
| 67 | STATIC void xfs_da3_node_rebalance(xfs_da_state_t *state, |
| 68 | xfs_da_state_blk_t *node_blk_1, |
| 69 | xfs_da_state_blk_t *node_blk_2); |
| 70 | STATIC void xfs_da3_node_add(xfs_da_state_t *state, |
| 71 | xfs_da_state_blk_t *old_node_blk, |
| 72 | xfs_da_state_blk_t *new_node_blk); |
| 73 | |
| 74 | /* |
| 75 | * Routines used for shrinking the Btree. |
| 76 | */ |
| 77 | STATIC int xfs_da3_root_join(xfs_da_state_t *state, |
| 78 | xfs_da_state_blk_t *root_blk); |
| 79 | STATIC int xfs_da3_node_toosmall(xfs_da_state_t *state, int *retval); |
| 80 | STATIC void xfs_da3_node_remove(xfs_da_state_t *state, |
| 81 | xfs_da_state_blk_t *drop_blk); |
| 82 | STATIC void xfs_da3_node_unbalance(xfs_da_state_t *state, |
| 83 | xfs_da_state_blk_t *src_node_blk, |
| 84 | xfs_da_state_blk_t *dst_node_blk); |
| 85 | |
| 86 | /* |
| 87 | * Utility routines. |
| 88 | */ |
| 89 | STATIC int xfs_da3_blk_unlink(xfs_da_state_t *state, |
| 90 | xfs_da_state_blk_t *drop_blk, |
| 91 | xfs_da_state_blk_t *save_blk); |
| 92 | |
| 93 | |
| 94 | kmem_zone_t *xfs_da_state_zone; /* anchor for state struct zone */ |
| 95 | |
| 96 | /* |
| 97 | * Allocate a dir-state structure. |
| 98 | * We don't put them on the stack since they're large. |
| 99 | */ |
| 100 | xfs_da_state_t * |
| 101 | xfs_da_state_alloc(void) |
| 102 | { |
| 103 | return kmem_zone_zalloc(xfs_da_state_zone, KM_NOFS); |
| 104 | } |
| 105 | |
| 106 | /* |
| 107 | * Kill the altpath contents of a da-state structure. |
| 108 | */ |
| 109 | STATIC void |
| 110 | xfs_da_state_kill_altpath(xfs_da_state_t *state) |
| 111 | { |
| 112 | int i; |
| 113 | |
| 114 | for (i = 0; i < state->altpath.active; i++) |
| 115 | state->altpath.blk[i].bp = NULL; |
| 116 | state->altpath.active = 0; |
| 117 | } |
| 118 | |
| 119 | /* |
| 120 | * Free a da-state structure. |
| 121 | */ |
| 122 | void |
| 123 | xfs_da_state_free(xfs_da_state_t *state) |
| 124 | { |
| 125 | xfs_da_state_kill_altpath(state); |
| 126 | #ifdef DEBUG |
| 127 | memset((char *)state, 0, sizeof(*state)); |
| 128 | #endif /* DEBUG */ |
| 129 | kmem_zone_free(xfs_da_state_zone, state); |
| 130 | } |
| 131 | |
| 132 | static bool |
| 133 | xfs_da3_node_verify( |
| 134 | struct xfs_buf *bp) |
| 135 | { |
| 136 | struct xfs_mount *mp = bp->b_target->bt_mount; |
| 137 | struct xfs_da_intnode *hdr = bp->b_addr; |
| 138 | struct xfs_da3_icnode_hdr ichdr; |
| 139 | const struct xfs_dir_ops *ops; |
| 140 | |
| 141 | ops = xfs_dir_get_ops(mp, NULL); |
| 142 | |
| 143 | ops->node_hdr_from_disk(&ichdr, hdr); |
| 144 | |
| 145 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
| 146 | struct xfs_da3_node_hdr *hdr3 = bp->b_addr; |
| 147 | |
| 148 | if (ichdr.magic != XFS_DA3_NODE_MAGIC) |
| 149 | return false; |
| 150 | |
| 151 | if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_uuid)) |
| 152 | return false; |
| 153 | if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn) |
| 154 | return false; |
| 155 | } else { |
| 156 | if (ichdr.magic != XFS_DA_NODE_MAGIC) |
| 157 | return false; |
| 158 | } |
| 159 | if (ichdr.level == 0) |
| 160 | return false; |
| 161 | if (ichdr.level > XFS_DA_NODE_MAXDEPTH) |
| 162 | return false; |
| 163 | if (ichdr.count == 0) |
| 164 | return false; |
| 165 | |
| 166 | /* |
| 167 | * we don't know if the node is for and attribute or directory tree, |
| 168 | * so only fail if the count is outside both bounds |
| 169 | */ |
| 170 | if (ichdr.count > mp->m_dir_node_ents && |
| 171 | ichdr.count > mp->m_attr_node_ents) |
| 172 | return false; |
| 173 | |
| 174 | /* XXX: hash order check? */ |
| 175 | |
| 176 | return true; |
| 177 | } |
| 178 | |
| 179 | static void |
| 180 | xfs_da3_node_write_verify( |
| 181 | struct xfs_buf *bp) |
| 182 | { |
| 183 | struct xfs_mount *mp = bp->b_target->bt_mount; |
| 184 | struct xfs_buf_log_item *bip = bp->b_fspriv; |
| 185 | struct xfs_da3_node_hdr *hdr3 = bp->b_addr; |
| 186 | |
| 187 | if (!xfs_da3_node_verify(bp)) { |
| 188 | xfs_buf_ioerror(bp, EFSCORRUPTED); |
| 189 | xfs_verifier_error(bp); |
| 190 | return; |
| 191 | } |
| 192 | |
| 193 | if (!xfs_sb_version_hascrc(&mp->m_sb)) |
| 194 | return; |
| 195 | |
| 196 | if (bip) |
| 197 | hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn); |
| 198 | |
| 199 | xfs_buf_update_cksum(bp, XFS_DA3_NODE_CRC_OFF); |
| 200 | } |
| 201 | |
| 202 | /* |
| 203 | * leaf/node format detection on trees is sketchy, so a node read can be done on |
| 204 | * leaf level blocks when detection identifies the tree as a node format tree |
| 205 | * incorrectly. In this case, we need to swap the verifier to match the correct |
| 206 | * format of the block being read. |
| 207 | */ |
| 208 | static void |
| 209 | xfs_da3_node_read_verify( |
| 210 | struct xfs_buf *bp) |
| 211 | { |
| 212 | struct xfs_da_blkinfo *info = bp->b_addr; |
| 213 | |
| 214 | switch (be16_to_cpu(info->magic)) { |
| 215 | case XFS_DA3_NODE_MAGIC: |
| 216 | if (!xfs_buf_verify_cksum(bp, XFS_DA3_NODE_CRC_OFF)) { |
| 217 | xfs_buf_ioerror(bp, EFSBADCRC); |
| 218 | break; |
| 219 | } |
| 220 | /* fall through */ |
| 221 | case XFS_DA_NODE_MAGIC: |
| 222 | if (!xfs_da3_node_verify(bp)) { |
| 223 | xfs_buf_ioerror(bp, EFSCORRUPTED); |
| 224 | break; |
| 225 | } |
| 226 | return; |
| 227 | case XFS_ATTR_LEAF_MAGIC: |
| 228 | case XFS_ATTR3_LEAF_MAGIC: |
| 229 | bp->b_ops = &xfs_attr3_leaf_buf_ops; |
| 230 | bp->b_ops->verify_read(bp); |
| 231 | return; |
| 232 | case XFS_DIR2_LEAFN_MAGIC: |
| 233 | case XFS_DIR3_LEAFN_MAGIC: |
| 234 | bp->b_ops = &xfs_dir3_leafn_buf_ops; |
| 235 | bp->b_ops->verify_read(bp); |
| 236 | return; |
| 237 | default: |
| 238 | break; |
| 239 | } |
| 240 | |
| 241 | /* corrupt block */ |
| 242 | xfs_verifier_error(bp); |
| 243 | } |
| 244 | |
| 245 | const struct xfs_buf_ops xfs_da3_node_buf_ops = { |
| 246 | .verify_read = xfs_da3_node_read_verify, |
| 247 | .verify_write = xfs_da3_node_write_verify, |
| 248 | }; |
| 249 | |
| 250 | int |
| 251 | xfs_da3_node_read( |
| 252 | struct xfs_trans *tp, |
| 253 | struct xfs_inode *dp, |
| 254 | xfs_dablk_t bno, |
| 255 | xfs_daddr_t mappedbno, |
| 256 | struct xfs_buf **bpp, |
| 257 | int which_fork) |
| 258 | { |
| 259 | int err; |
| 260 | |
| 261 | err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp, |
| 262 | which_fork, &xfs_da3_node_buf_ops); |
| 263 | if (!err && tp) { |
| 264 | struct xfs_da_blkinfo *info = (*bpp)->b_addr; |
| 265 | int type; |
| 266 | |
| 267 | switch (be16_to_cpu(info->magic)) { |
| 268 | case XFS_DA_NODE_MAGIC: |
| 269 | case XFS_DA3_NODE_MAGIC: |
| 270 | type = XFS_BLFT_DA_NODE_BUF; |
| 271 | break; |
| 272 | case XFS_ATTR_LEAF_MAGIC: |
| 273 | case XFS_ATTR3_LEAF_MAGIC: |
| 274 | type = XFS_BLFT_ATTR_LEAF_BUF; |
| 275 | break; |
| 276 | case XFS_DIR2_LEAFN_MAGIC: |
| 277 | case XFS_DIR3_LEAFN_MAGIC: |
| 278 | type = XFS_BLFT_DIR_LEAFN_BUF; |
| 279 | break; |
| 280 | default: |
| 281 | type = 0; |
| 282 | ASSERT(0); |
| 283 | break; |
| 284 | } |
| 285 | xfs_trans_buf_set_type(tp, *bpp, type); |
| 286 | } |
| 287 | return err; |
| 288 | } |
| 289 | |
| 290 | /*======================================================================== |
| 291 | * Routines used for growing the Btree. |
| 292 | *========================================================================*/ |
| 293 | |
| 294 | /* |
| 295 | * Create the initial contents of an intermediate node. |
| 296 | */ |
| 297 | int |
| 298 | xfs_da3_node_create( |
| 299 | struct xfs_da_args *args, |
| 300 | xfs_dablk_t blkno, |
| 301 | int level, |
| 302 | struct xfs_buf **bpp, |
| 303 | int whichfork) |
| 304 | { |
| 305 | struct xfs_da_intnode *node; |
| 306 | struct xfs_trans *tp = args->trans; |
| 307 | struct xfs_mount *mp = tp->t_mountp; |
| 308 | struct xfs_da3_icnode_hdr ichdr = {0}; |
| 309 | struct xfs_buf *bp; |
| 310 | int error; |
| 311 | struct xfs_inode *dp = args->dp; |
| 312 | |
| 313 | trace_xfs_da_node_create(args); |
| 314 | ASSERT(level <= XFS_DA_NODE_MAXDEPTH); |
| 315 | |
| 316 | error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, whichfork); |
| 317 | if (error) |
| 318 | return(error); |
| 319 | bp->b_ops = &xfs_da3_node_buf_ops; |
| 320 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF); |
| 321 | node = bp->b_addr; |
| 322 | |
| 323 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
| 324 | struct xfs_da3_node_hdr *hdr3 = bp->b_addr; |
| 325 | |
| 326 | ichdr.magic = XFS_DA3_NODE_MAGIC; |
| 327 | hdr3->info.blkno = cpu_to_be64(bp->b_bn); |
| 328 | hdr3->info.owner = cpu_to_be64(args->dp->i_ino); |
| 329 | uuid_copy(&hdr3->info.uuid, &mp->m_sb.sb_uuid); |
| 330 | } else { |
| 331 | ichdr.magic = XFS_DA_NODE_MAGIC; |
| 332 | } |
| 333 | ichdr.level = level; |
| 334 | |
| 335 | dp->d_ops->node_hdr_to_disk(node, &ichdr); |
| 336 | xfs_trans_log_buf(tp, bp, |
| 337 | XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size)); |
| 338 | |
| 339 | *bpp = bp; |
| 340 | return(0); |
| 341 | } |
| 342 | |
| 343 | /* |
| 344 | * Split a leaf node, rebalance, then possibly split |
| 345 | * intermediate nodes, rebalance, etc. |
| 346 | */ |
| 347 | int /* error */ |
| 348 | xfs_da3_split( |
| 349 | struct xfs_da_state *state) |
| 350 | { |
| 351 | struct xfs_da_state_blk *oldblk; |
| 352 | struct xfs_da_state_blk *newblk; |
| 353 | struct xfs_da_state_blk *addblk; |
| 354 | struct xfs_da_intnode *node; |
| 355 | struct xfs_buf *bp; |
| 356 | int max; |
| 357 | int action = 0; |
| 358 | int error; |
| 359 | int i; |
| 360 | |
| 361 | trace_xfs_da_split(state->args); |
| 362 | |
| 363 | /* |
| 364 | * Walk back up the tree splitting/inserting/adjusting as necessary. |
| 365 | * If we need to insert and there isn't room, split the node, then |
| 366 | * decide which fragment to insert the new block from below into. |
| 367 | * Note that we may split the root this way, but we need more fixup. |
| 368 | */ |
| 369 | max = state->path.active - 1; |
| 370 | ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH)); |
| 371 | ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC || |
| 372 | state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC); |
| 373 | |
| 374 | addblk = &state->path.blk[max]; /* initial dummy value */ |
| 375 | for (i = max; (i >= 0) && addblk; state->path.active--, i--) { |
| 376 | oldblk = &state->path.blk[i]; |
| 377 | newblk = &state->altpath.blk[i]; |
| 378 | |
| 379 | /* |
| 380 | * If a leaf node then |
| 381 | * Allocate a new leaf node, then rebalance across them. |
| 382 | * else if an intermediate node then |
| 383 | * We split on the last layer, must we split the node? |
| 384 | */ |
| 385 | switch (oldblk->magic) { |
| 386 | case XFS_ATTR_LEAF_MAGIC: |
| 387 | error = xfs_attr3_leaf_split(state, oldblk, newblk); |
| 388 | if ((error != 0) && (error != ENOSPC)) { |
| 389 | return(error); /* GROT: attr is inconsistent */ |
| 390 | } |
| 391 | if (!error) { |
| 392 | addblk = newblk; |
| 393 | break; |
| 394 | } |
| 395 | /* |
| 396 | * Entry wouldn't fit, split the leaf again. |
| 397 | */ |
| 398 | state->extravalid = 1; |
| 399 | if (state->inleaf) { |
| 400 | state->extraafter = 0; /* before newblk */ |
| 401 | trace_xfs_attr_leaf_split_before(state->args); |
| 402 | error = xfs_attr3_leaf_split(state, oldblk, |
| 403 | &state->extrablk); |
| 404 | } else { |
| 405 | state->extraafter = 1; /* after newblk */ |
| 406 | trace_xfs_attr_leaf_split_after(state->args); |
| 407 | error = xfs_attr3_leaf_split(state, newblk, |
| 408 | &state->extrablk); |
| 409 | } |
| 410 | if (error) |
| 411 | return(error); /* GROT: attr inconsistent */ |
| 412 | addblk = newblk; |
| 413 | break; |
| 414 | case XFS_DIR2_LEAFN_MAGIC: |
| 415 | error = xfs_dir2_leafn_split(state, oldblk, newblk); |
| 416 | if (error) |
| 417 | return error; |
| 418 | addblk = newblk; |
| 419 | break; |
| 420 | case XFS_DA_NODE_MAGIC: |
| 421 | error = xfs_da3_node_split(state, oldblk, newblk, addblk, |
| 422 | max - i, &action); |
| 423 | addblk->bp = NULL; |
| 424 | if (error) |
| 425 | return(error); /* GROT: dir is inconsistent */ |
| 426 | /* |
| 427 | * Record the newly split block for the next time thru? |
| 428 | */ |
| 429 | if (action) |
| 430 | addblk = newblk; |
| 431 | else |
| 432 | addblk = NULL; |
| 433 | break; |
| 434 | } |
| 435 | |
| 436 | /* |
| 437 | * Update the btree to show the new hashval for this child. |
| 438 | */ |
| 439 | xfs_da3_fixhashpath(state, &state->path); |
| 440 | } |
| 441 | if (!addblk) |
| 442 | return(0); |
| 443 | |
| 444 | /* |
| 445 | * Split the root node. |
| 446 | */ |
| 447 | ASSERT(state->path.active == 0); |
| 448 | oldblk = &state->path.blk[0]; |
| 449 | error = xfs_da3_root_split(state, oldblk, addblk); |
| 450 | if (error) { |
| 451 | addblk->bp = NULL; |
| 452 | return(error); /* GROT: dir is inconsistent */ |
| 453 | } |
| 454 | |
| 455 | /* |
| 456 | * Update pointers to the node which used to be block 0 and |
| 457 | * just got bumped because of the addition of a new root node. |
| 458 | * There might be three blocks involved if a double split occurred, |
| 459 | * and the original block 0 could be at any position in the list. |
| 460 | * |
| 461 | * Note: the magic numbers and sibling pointers are in the same |
| 462 | * physical place for both v2 and v3 headers (by design). Hence it |
| 463 | * doesn't matter which version of the xfs_da_intnode structure we use |
| 464 | * here as the result will be the same using either structure. |
| 465 | */ |
| 466 | node = oldblk->bp->b_addr; |
| 467 | if (node->hdr.info.forw) { |
| 468 | if (be32_to_cpu(node->hdr.info.forw) == addblk->blkno) { |
| 469 | bp = addblk->bp; |
| 470 | } else { |
| 471 | ASSERT(state->extravalid); |
| 472 | bp = state->extrablk.bp; |
| 473 | } |
| 474 | node = bp->b_addr; |
| 475 | node->hdr.info.back = cpu_to_be32(oldblk->blkno); |
| 476 | xfs_trans_log_buf(state->args->trans, bp, |
| 477 | XFS_DA_LOGRANGE(node, &node->hdr.info, |
| 478 | sizeof(node->hdr.info))); |
| 479 | } |
| 480 | node = oldblk->bp->b_addr; |
| 481 | if (node->hdr.info.back) { |
| 482 | if (be32_to_cpu(node->hdr.info.back) == addblk->blkno) { |
| 483 | bp = addblk->bp; |
| 484 | } else { |
| 485 | ASSERT(state->extravalid); |
| 486 | bp = state->extrablk.bp; |
| 487 | } |
| 488 | node = bp->b_addr; |
| 489 | node->hdr.info.forw = cpu_to_be32(oldblk->blkno); |
| 490 | xfs_trans_log_buf(state->args->trans, bp, |
| 491 | XFS_DA_LOGRANGE(node, &node->hdr.info, |
| 492 | sizeof(node->hdr.info))); |
| 493 | } |
| 494 | addblk->bp = NULL; |
| 495 | return(0); |
| 496 | } |
| 497 | |
| 498 | /* |
| 499 | * Split the root. We have to create a new root and point to the two |
| 500 | * parts (the split old root) that we just created. Copy block zero to |
| 501 | * the EOF, extending the inode in process. |
| 502 | */ |
| 503 | STATIC int /* error */ |
| 504 | xfs_da3_root_split( |
| 505 | struct xfs_da_state *state, |
| 506 | struct xfs_da_state_blk *blk1, |
| 507 | struct xfs_da_state_blk *blk2) |
| 508 | { |
| 509 | struct xfs_da_intnode *node; |
| 510 | struct xfs_da_intnode *oldroot; |
| 511 | struct xfs_da_node_entry *btree; |
| 512 | struct xfs_da3_icnode_hdr nodehdr; |
| 513 | struct xfs_da_args *args; |
| 514 | struct xfs_buf *bp; |
| 515 | struct xfs_inode *dp; |
| 516 | struct xfs_trans *tp; |
| 517 | struct xfs_mount *mp; |
| 518 | struct xfs_dir2_leaf *leaf; |
| 519 | xfs_dablk_t blkno; |
| 520 | int level; |
| 521 | int error; |
| 522 | int size; |
| 523 | |
| 524 | trace_xfs_da_root_split(state->args); |
| 525 | |
| 526 | /* |
| 527 | * Copy the existing (incorrect) block from the root node position |
| 528 | * to a free space somewhere. |
| 529 | */ |
| 530 | args = state->args; |
| 531 | error = xfs_da_grow_inode(args, &blkno); |
| 532 | if (error) |
| 533 | return error; |
| 534 | |
| 535 | dp = args->dp; |
| 536 | tp = args->trans; |
| 537 | mp = state->mp; |
| 538 | error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork); |
| 539 | if (error) |
| 540 | return error; |
| 541 | node = bp->b_addr; |
| 542 | oldroot = blk1->bp->b_addr; |
| 543 | if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC) || |
| 544 | oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC)) { |
| 545 | struct xfs_da3_icnode_hdr nodehdr; |
| 546 | |
| 547 | dp->d_ops->node_hdr_from_disk(&nodehdr, oldroot); |
| 548 | btree = dp->d_ops->node_tree_p(oldroot); |
| 549 | size = (int)((char *)&btree[nodehdr.count] - (char *)oldroot); |
| 550 | level = nodehdr.level; |
| 551 | |
| 552 | /* |
| 553 | * we are about to copy oldroot to bp, so set up the type |
| 554 | * of bp while we know exactly what it will be. |
| 555 | */ |
| 556 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF); |
| 557 | } else { |
| 558 | struct xfs_dir3_icleaf_hdr leafhdr; |
| 559 | struct xfs_dir2_leaf_entry *ents; |
| 560 | |
| 561 | leaf = (xfs_dir2_leaf_t *)oldroot; |
| 562 | dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf); |
| 563 | ents = dp->d_ops->leaf_ents_p(leaf); |
| 564 | |
| 565 | ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC || |
| 566 | leafhdr.magic == XFS_DIR3_LEAFN_MAGIC); |
| 567 | size = (int)((char *)&ents[leafhdr.count] - (char *)leaf); |
| 568 | level = 0; |
| 569 | |
| 570 | /* |
| 571 | * we are about to copy oldroot to bp, so set up the type |
| 572 | * of bp while we know exactly what it will be. |
| 573 | */ |
| 574 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF); |
| 575 | } |
| 576 | |
| 577 | /* |
| 578 | * we can copy most of the information in the node from one block to |
| 579 | * another, but for CRC enabled headers we have to make sure that the |
| 580 | * block specific identifiers are kept intact. We update the buffer |
| 581 | * directly for this. |
| 582 | */ |
| 583 | memcpy(node, oldroot, size); |
| 584 | if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) || |
| 585 | oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) { |
| 586 | struct xfs_da3_intnode *node3 = (struct xfs_da3_intnode *)node; |
| 587 | |
| 588 | node3->hdr.info.blkno = cpu_to_be64(bp->b_bn); |
| 589 | } |
| 590 | xfs_trans_log_buf(tp, bp, 0, size - 1); |
| 591 | |
| 592 | bp->b_ops = blk1->bp->b_ops; |
| 593 | xfs_trans_buf_copy_type(bp, blk1->bp); |
| 594 | blk1->bp = bp; |
| 595 | blk1->blkno = blkno; |
| 596 | |
| 597 | /* |
| 598 | * Set up the new root node. |
| 599 | */ |
| 600 | error = xfs_da3_node_create(args, |
| 601 | (args->whichfork == XFS_DATA_FORK) ? args->geo->leafblk : 0, |
| 602 | level + 1, &bp, args->whichfork); |
| 603 | if (error) |
| 604 | return error; |
| 605 | |
| 606 | node = bp->b_addr; |
| 607 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
| 608 | btree = dp->d_ops->node_tree_p(node); |
| 609 | btree[0].hashval = cpu_to_be32(blk1->hashval); |
| 610 | btree[0].before = cpu_to_be32(blk1->blkno); |
| 611 | btree[1].hashval = cpu_to_be32(blk2->hashval); |
| 612 | btree[1].before = cpu_to_be32(blk2->blkno); |
| 613 | nodehdr.count = 2; |
| 614 | dp->d_ops->node_hdr_to_disk(node, &nodehdr); |
| 615 | |
| 616 | #ifdef DEBUG |
| 617 | if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) || |
| 618 | oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) { |
| 619 | ASSERT(blk1->blkno >= args->geo->leafblk && |
| 620 | blk1->blkno < args->geo->freeblk); |
| 621 | ASSERT(blk2->blkno >= args->geo->leafblk && |
| 622 | blk2->blkno < args->geo->freeblk); |
| 623 | } |
| 624 | #endif |
| 625 | |
| 626 | /* Header is already logged by xfs_da_node_create */ |
| 627 | xfs_trans_log_buf(tp, bp, |
| 628 | XFS_DA_LOGRANGE(node, btree, sizeof(xfs_da_node_entry_t) * 2)); |
| 629 | |
| 630 | return 0; |
| 631 | } |
| 632 | |
| 633 | /* |
| 634 | * Split the node, rebalance, then add the new entry. |
| 635 | */ |
| 636 | STATIC int /* error */ |
| 637 | xfs_da3_node_split( |
| 638 | struct xfs_da_state *state, |
| 639 | struct xfs_da_state_blk *oldblk, |
| 640 | struct xfs_da_state_blk *newblk, |
| 641 | struct xfs_da_state_blk *addblk, |
| 642 | int treelevel, |
| 643 | int *result) |
| 644 | { |
| 645 | struct xfs_da_intnode *node; |
| 646 | struct xfs_da3_icnode_hdr nodehdr; |
| 647 | xfs_dablk_t blkno; |
| 648 | int newcount; |
| 649 | int error; |
| 650 | int useextra; |
| 651 | struct xfs_inode *dp = state->args->dp; |
| 652 | |
| 653 | trace_xfs_da_node_split(state->args); |
| 654 | |
| 655 | node = oldblk->bp->b_addr; |
| 656 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
| 657 | |
| 658 | /* |
| 659 | * With V2 dirs the extra block is data or freespace. |
| 660 | */ |
| 661 | useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK; |
| 662 | newcount = 1 + useextra; |
| 663 | /* |
| 664 | * Do we have to split the node? |
| 665 | */ |
| 666 | if (nodehdr.count + newcount > state->node_ents) { |
| 667 | /* |
| 668 | * Allocate a new node, add to the doubly linked chain of |
| 669 | * nodes, then move some of our excess entries into it. |
| 670 | */ |
| 671 | error = xfs_da_grow_inode(state->args, &blkno); |
| 672 | if (error) |
| 673 | return(error); /* GROT: dir is inconsistent */ |
| 674 | |
| 675 | error = xfs_da3_node_create(state->args, blkno, treelevel, |
| 676 | &newblk->bp, state->args->whichfork); |
| 677 | if (error) |
| 678 | return(error); /* GROT: dir is inconsistent */ |
| 679 | newblk->blkno = blkno; |
| 680 | newblk->magic = XFS_DA_NODE_MAGIC; |
| 681 | xfs_da3_node_rebalance(state, oldblk, newblk); |
| 682 | error = xfs_da3_blk_link(state, oldblk, newblk); |
| 683 | if (error) |
| 684 | return(error); |
| 685 | *result = 1; |
| 686 | } else { |
| 687 | *result = 0; |
| 688 | } |
| 689 | |
| 690 | /* |
| 691 | * Insert the new entry(s) into the correct block |
| 692 | * (updating last hashval in the process). |
| 693 | * |
| 694 | * xfs_da3_node_add() inserts BEFORE the given index, |
| 695 | * and as a result of using node_lookup_int() we always |
| 696 | * point to a valid entry (not after one), but a split |
| 697 | * operation always results in a new block whose hashvals |
| 698 | * FOLLOW the current block. |
| 699 | * |
| 700 | * If we had double-split op below us, then add the extra block too. |
| 701 | */ |
| 702 | node = oldblk->bp->b_addr; |
| 703 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
| 704 | if (oldblk->index <= nodehdr.count) { |
| 705 | oldblk->index++; |
| 706 | xfs_da3_node_add(state, oldblk, addblk); |
| 707 | if (useextra) { |
| 708 | if (state->extraafter) |
| 709 | oldblk->index++; |
| 710 | xfs_da3_node_add(state, oldblk, &state->extrablk); |
| 711 | state->extravalid = 0; |
| 712 | } |
| 713 | } else { |
| 714 | newblk->index++; |
| 715 | xfs_da3_node_add(state, newblk, addblk); |
| 716 | if (useextra) { |
| 717 | if (state->extraafter) |
| 718 | newblk->index++; |
| 719 | xfs_da3_node_add(state, newblk, &state->extrablk); |
| 720 | state->extravalid = 0; |
| 721 | } |
| 722 | } |
| 723 | |
| 724 | return(0); |
| 725 | } |
| 726 | |
| 727 | /* |
| 728 | * Balance the btree elements between two intermediate nodes, |
| 729 | * usually one full and one empty. |
| 730 | * |
| 731 | * NOTE: if blk2 is empty, then it will get the upper half of blk1. |
| 732 | */ |
| 733 | STATIC void |
| 734 | xfs_da3_node_rebalance( |
| 735 | struct xfs_da_state *state, |
| 736 | struct xfs_da_state_blk *blk1, |
| 737 | struct xfs_da_state_blk *blk2) |
| 738 | { |
| 739 | struct xfs_da_intnode *node1; |
| 740 | struct xfs_da_intnode *node2; |
| 741 | struct xfs_da_intnode *tmpnode; |
| 742 | struct xfs_da_node_entry *btree1; |
| 743 | struct xfs_da_node_entry *btree2; |
| 744 | struct xfs_da_node_entry *btree_s; |
| 745 | struct xfs_da_node_entry *btree_d; |
| 746 | struct xfs_da3_icnode_hdr nodehdr1; |
| 747 | struct xfs_da3_icnode_hdr nodehdr2; |
| 748 | struct xfs_trans *tp; |
| 749 | int count; |
| 750 | int tmp; |
| 751 | int swap = 0; |
| 752 | struct xfs_inode *dp = state->args->dp; |
| 753 | |
| 754 | trace_xfs_da_node_rebalance(state->args); |
| 755 | |
| 756 | node1 = blk1->bp->b_addr; |
| 757 | node2 = blk2->bp->b_addr; |
| 758 | dp->d_ops->node_hdr_from_disk(&nodehdr1, node1); |
| 759 | dp->d_ops->node_hdr_from_disk(&nodehdr2, node2); |
| 760 | btree1 = dp->d_ops->node_tree_p(node1); |
| 761 | btree2 = dp->d_ops->node_tree_p(node2); |
| 762 | |
| 763 | /* |
| 764 | * Figure out how many entries need to move, and in which direction. |
| 765 | * Swap the nodes around if that makes it simpler. |
| 766 | */ |
| 767 | if (nodehdr1.count > 0 && nodehdr2.count > 0 && |
| 768 | ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) || |
| 769 | (be32_to_cpu(btree2[nodehdr2.count - 1].hashval) < |
| 770 | be32_to_cpu(btree1[nodehdr1.count - 1].hashval)))) { |
| 771 | tmpnode = node1; |
| 772 | node1 = node2; |
| 773 | node2 = tmpnode; |
| 774 | dp->d_ops->node_hdr_from_disk(&nodehdr1, node1); |
| 775 | dp->d_ops->node_hdr_from_disk(&nodehdr2, node2); |
| 776 | btree1 = dp->d_ops->node_tree_p(node1); |
| 777 | btree2 = dp->d_ops->node_tree_p(node2); |
| 778 | swap = 1; |
| 779 | } |
| 780 | |
| 781 | count = (nodehdr1.count - nodehdr2.count) / 2; |
| 782 | if (count == 0) |
| 783 | return; |
| 784 | tp = state->args->trans; |
| 785 | /* |
| 786 | * Two cases: high-to-low and low-to-high. |
| 787 | */ |
| 788 | if (count > 0) { |
| 789 | /* |
| 790 | * Move elements in node2 up to make a hole. |
| 791 | */ |
| 792 | tmp = nodehdr2.count; |
| 793 | if (tmp > 0) { |
| 794 | tmp *= (uint)sizeof(xfs_da_node_entry_t); |
| 795 | btree_s = &btree2[0]; |
| 796 | btree_d = &btree2[count]; |
| 797 | memmove(btree_d, btree_s, tmp); |
| 798 | } |
| 799 | |
| 800 | /* |
| 801 | * Move the req'd B-tree elements from high in node1 to |
| 802 | * low in node2. |
| 803 | */ |
| 804 | nodehdr2.count += count; |
| 805 | tmp = count * (uint)sizeof(xfs_da_node_entry_t); |
| 806 | btree_s = &btree1[nodehdr1.count - count]; |
| 807 | btree_d = &btree2[0]; |
| 808 | memcpy(btree_d, btree_s, tmp); |
| 809 | nodehdr1.count -= count; |
| 810 | } else { |
| 811 | /* |
| 812 | * Move the req'd B-tree elements from low in node2 to |
| 813 | * high in node1. |
| 814 | */ |
| 815 | count = -count; |
| 816 | tmp = count * (uint)sizeof(xfs_da_node_entry_t); |
| 817 | btree_s = &btree2[0]; |
| 818 | btree_d = &btree1[nodehdr1.count]; |
| 819 | memcpy(btree_d, btree_s, tmp); |
| 820 | nodehdr1.count += count; |
| 821 | |
| 822 | xfs_trans_log_buf(tp, blk1->bp, |
| 823 | XFS_DA_LOGRANGE(node1, btree_d, tmp)); |
| 824 | |
| 825 | /* |
| 826 | * Move elements in node2 down to fill the hole. |
| 827 | */ |
| 828 | tmp = nodehdr2.count - count; |
| 829 | tmp *= (uint)sizeof(xfs_da_node_entry_t); |
| 830 | btree_s = &btree2[count]; |
| 831 | btree_d = &btree2[0]; |
| 832 | memmove(btree_d, btree_s, tmp); |
| 833 | nodehdr2.count -= count; |
| 834 | } |
| 835 | |
| 836 | /* |
| 837 | * Log header of node 1 and all current bits of node 2. |
| 838 | */ |
| 839 | dp->d_ops->node_hdr_to_disk(node1, &nodehdr1); |
| 840 | xfs_trans_log_buf(tp, blk1->bp, |
| 841 | XFS_DA_LOGRANGE(node1, &node1->hdr, dp->d_ops->node_hdr_size)); |
| 842 | |
| 843 | dp->d_ops->node_hdr_to_disk(node2, &nodehdr2); |
| 844 | xfs_trans_log_buf(tp, blk2->bp, |
| 845 | XFS_DA_LOGRANGE(node2, &node2->hdr, |
| 846 | dp->d_ops->node_hdr_size + |
| 847 | (sizeof(btree2[0]) * nodehdr2.count))); |
| 848 | |
| 849 | /* |
| 850 | * Record the last hashval from each block for upward propagation. |
| 851 | * (note: don't use the swapped node pointers) |
| 852 | */ |
| 853 | if (swap) { |
| 854 | node1 = blk1->bp->b_addr; |
| 855 | node2 = blk2->bp->b_addr; |
| 856 | dp->d_ops->node_hdr_from_disk(&nodehdr1, node1); |
| 857 | dp->d_ops->node_hdr_from_disk(&nodehdr2, node2); |
| 858 | btree1 = dp->d_ops->node_tree_p(node1); |
| 859 | btree2 = dp->d_ops->node_tree_p(node2); |
| 860 | } |
| 861 | blk1->hashval = be32_to_cpu(btree1[nodehdr1.count - 1].hashval); |
| 862 | blk2->hashval = be32_to_cpu(btree2[nodehdr2.count - 1].hashval); |
| 863 | |
| 864 | /* |
| 865 | * Adjust the expected index for insertion. |
| 866 | */ |
| 867 | if (blk1->index >= nodehdr1.count) { |
| 868 | blk2->index = blk1->index - nodehdr1.count; |
| 869 | blk1->index = nodehdr1.count + 1; /* make it invalid */ |
| 870 | } |
| 871 | } |
| 872 | |
| 873 | /* |
| 874 | * Add a new entry to an intermediate node. |
| 875 | */ |
| 876 | STATIC void |
| 877 | xfs_da3_node_add( |
| 878 | struct xfs_da_state *state, |
| 879 | struct xfs_da_state_blk *oldblk, |
| 880 | struct xfs_da_state_blk *newblk) |
| 881 | { |
| 882 | struct xfs_da_intnode *node; |
| 883 | struct xfs_da3_icnode_hdr nodehdr; |
| 884 | struct xfs_da_node_entry *btree; |
| 885 | int tmp; |
| 886 | struct xfs_inode *dp = state->args->dp; |
| 887 | |
| 888 | trace_xfs_da_node_add(state->args); |
| 889 | |
| 890 | node = oldblk->bp->b_addr; |
| 891 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
| 892 | btree = dp->d_ops->node_tree_p(node); |
| 893 | |
| 894 | ASSERT(oldblk->index >= 0 && oldblk->index <= nodehdr.count); |
| 895 | ASSERT(newblk->blkno != 0); |
| 896 | if (state->args->whichfork == XFS_DATA_FORK) |
| 897 | ASSERT(newblk->blkno >= state->args->geo->leafblk && |
| 898 | newblk->blkno < state->args->geo->freeblk); |
| 899 | |
| 900 | /* |
| 901 | * We may need to make some room before we insert the new node. |
| 902 | */ |
| 903 | tmp = 0; |
| 904 | if (oldblk->index < nodehdr.count) { |
| 905 | tmp = (nodehdr.count - oldblk->index) * (uint)sizeof(*btree); |
| 906 | memmove(&btree[oldblk->index + 1], &btree[oldblk->index], tmp); |
| 907 | } |
| 908 | btree[oldblk->index].hashval = cpu_to_be32(newblk->hashval); |
| 909 | btree[oldblk->index].before = cpu_to_be32(newblk->blkno); |
| 910 | xfs_trans_log_buf(state->args->trans, oldblk->bp, |
| 911 | XFS_DA_LOGRANGE(node, &btree[oldblk->index], |
| 912 | tmp + sizeof(*btree))); |
| 913 | |
| 914 | nodehdr.count += 1; |
| 915 | dp->d_ops->node_hdr_to_disk(node, &nodehdr); |
| 916 | xfs_trans_log_buf(state->args->trans, oldblk->bp, |
| 917 | XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size)); |
| 918 | |
| 919 | /* |
| 920 | * Copy the last hash value from the oldblk to propagate upwards. |
| 921 | */ |
| 922 | oldblk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval); |
| 923 | } |
| 924 | |
| 925 | /*======================================================================== |
| 926 | * Routines used for shrinking the Btree. |
| 927 | *========================================================================*/ |
| 928 | |
| 929 | /* |
| 930 | * Deallocate an empty leaf node, remove it from its parent, |
| 931 | * possibly deallocating that block, etc... |
| 932 | */ |
| 933 | int |
| 934 | xfs_da3_join( |
| 935 | struct xfs_da_state *state) |
| 936 | { |
| 937 | struct xfs_da_state_blk *drop_blk; |
| 938 | struct xfs_da_state_blk *save_blk; |
| 939 | int action = 0; |
| 940 | int error; |
| 941 | |
| 942 | trace_xfs_da_join(state->args); |
| 943 | |
| 944 | drop_blk = &state->path.blk[ state->path.active-1 ]; |
| 945 | save_blk = &state->altpath.blk[ state->path.active-1 ]; |
| 946 | ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC); |
| 947 | ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC || |
| 948 | drop_blk->magic == XFS_DIR2_LEAFN_MAGIC); |
| 949 | |
| 950 | /* |
| 951 | * Walk back up the tree joining/deallocating as necessary. |
| 952 | * When we stop dropping blocks, break out. |
| 953 | */ |
| 954 | for ( ; state->path.active >= 2; drop_blk--, save_blk--, |
| 955 | state->path.active--) { |
| 956 | /* |
| 957 | * See if we can combine the block with a neighbor. |
| 958 | * (action == 0) => no options, just leave |
| 959 | * (action == 1) => coalesce, then unlink |
| 960 | * (action == 2) => block empty, unlink it |
| 961 | */ |
| 962 | switch (drop_blk->magic) { |
| 963 | case XFS_ATTR_LEAF_MAGIC: |
| 964 | error = xfs_attr3_leaf_toosmall(state, &action); |
| 965 | if (error) |
| 966 | return(error); |
| 967 | if (action == 0) |
| 968 | return(0); |
| 969 | xfs_attr3_leaf_unbalance(state, drop_blk, save_blk); |
| 970 | break; |
| 971 | case XFS_DIR2_LEAFN_MAGIC: |
| 972 | error = xfs_dir2_leafn_toosmall(state, &action); |
| 973 | if (error) |
| 974 | return error; |
| 975 | if (action == 0) |
| 976 | return 0; |
| 977 | xfs_dir2_leafn_unbalance(state, drop_blk, save_blk); |
| 978 | break; |
| 979 | case XFS_DA_NODE_MAGIC: |
| 980 | /* |
| 981 | * Remove the offending node, fixup hashvals, |
| 982 | * check for a toosmall neighbor. |
| 983 | */ |
| 984 | xfs_da3_node_remove(state, drop_blk); |
| 985 | xfs_da3_fixhashpath(state, &state->path); |
| 986 | error = xfs_da3_node_toosmall(state, &action); |
| 987 | if (error) |
| 988 | return(error); |
| 989 | if (action == 0) |
| 990 | return 0; |
| 991 | xfs_da3_node_unbalance(state, drop_blk, save_blk); |
| 992 | break; |
| 993 | } |
| 994 | xfs_da3_fixhashpath(state, &state->altpath); |
| 995 | error = xfs_da3_blk_unlink(state, drop_blk, save_blk); |
| 996 | xfs_da_state_kill_altpath(state); |
| 997 | if (error) |
| 998 | return(error); |
| 999 | error = xfs_da_shrink_inode(state->args, drop_blk->blkno, |
| 1000 | drop_blk->bp); |
| 1001 | drop_blk->bp = NULL; |
| 1002 | if (error) |
| 1003 | return(error); |
| 1004 | } |
| 1005 | /* |
| 1006 | * We joined all the way to the top. If it turns out that |
| 1007 | * we only have one entry in the root, make the child block |
| 1008 | * the new root. |
| 1009 | */ |
| 1010 | xfs_da3_node_remove(state, drop_blk); |
| 1011 | xfs_da3_fixhashpath(state, &state->path); |
| 1012 | error = xfs_da3_root_join(state, &state->path.blk[0]); |
| 1013 | return(error); |
| 1014 | } |
| 1015 | |
| 1016 | #ifdef DEBUG |
| 1017 | static void |
| 1018 | xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo *blkinfo, __u16 level) |
| 1019 | { |
| 1020 | __be16 magic = blkinfo->magic; |
| 1021 | |
| 1022 | if (level == 1) { |
| 1023 | ASSERT(magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) || |
| 1024 | magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) || |
| 1025 | magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) || |
| 1026 | magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)); |
| 1027 | } else { |
| 1028 | ASSERT(magic == cpu_to_be16(XFS_DA_NODE_MAGIC) || |
| 1029 | magic == cpu_to_be16(XFS_DA3_NODE_MAGIC)); |
| 1030 | } |
| 1031 | ASSERT(!blkinfo->forw); |
| 1032 | ASSERT(!blkinfo->back); |
| 1033 | } |
| 1034 | #else /* !DEBUG */ |
| 1035 | #define xfs_da_blkinfo_onlychild_validate(blkinfo, level) |
| 1036 | #endif /* !DEBUG */ |
| 1037 | |
| 1038 | /* |
| 1039 | * We have only one entry in the root. Copy the only remaining child of |
| 1040 | * the old root to block 0 as the new root node. |
| 1041 | */ |
| 1042 | STATIC int |
| 1043 | xfs_da3_root_join( |
| 1044 | struct xfs_da_state *state, |
| 1045 | struct xfs_da_state_blk *root_blk) |
| 1046 | { |
| 1047 | struct xfs_da_intnode *oldroot; |
| 1048 | struct xfs_da_args *args; |
| 1049 | xfs_dablk_t child; |
| 1050 | struct xfs_buf *bp; |
| 1051 | struct xfs_da3_icnode_hdr oldroothdr; |
| 1052 | struct xfs_da_node_entry *btree; |
| 1053 | int error; |
| 1054 | struct xfs_inode *dp = state->args->dp; |
| 1055 | |
| 1056 | trace_xfs_da_root_join(state->args); |
| 1057 | |
| 1058 | ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC); |
| 1059 | |
| 1060 | args = state->args; |
| 1061 | oldroot = root_blk->bp->b_addr; |
| 1062 | dp->d_ops->node_hdr_from_disk(&oldroothdr, oldroot); |
| 1063 | ASSERT(oldroothdr.forw == 0); |
| 1064 | ASSERT(oldroothdr.back == 0); |
| 1065 | |
| 1066 | /* |
| 1067 | * If the root has more than one child, then don't do anything. |
| 1068 | */ |
| 1069 | if (oldroothdr.count > 1) |
| 1070 | return 0; |
| 1071 | |
| 1072 | /* |
| 1073 | * Read in the (only) child block, then copy those bytes into |
| 1074 | * the root block's buffer and free the original child block. |
| 1075 | */ |
| 1076 | btree = dp->d_ops->node_tree_p(oldroot); |
| 1077 | child = be32_to_cpu(btree[0].before); |
| 1078 | ASSERT(child != 0); |
| 1079 | error = xfs_da3_node_read(args->trans, dp, child, -1, &bp, |
| 1080 | args->whichfork); |
| 1081 | if (error) |
| 1082 | return error; |
| 1083 | xfs_da_blkinfo_onlychild_validate(bp->b_addr, oldroothdr.level); |
| 1084 | |
| 1085 | /* |
| 1086 | * This could be copying a leaf back into the root block in the case of |
| 1087 | * there only being a single leaf block left in the tree. Hence we have |
| 1088 | * to update the b_ops pointer as well to match the buffer type change |
| 1089 | * that could occur. For dir3 blocks we also need to update the block |
| 1090 | * number in the buffer header. |
| 1091 | */ |
| 1092 | memcpy(root_blk->bp->b_addr, bp->b_addr, state->blocksize); |
| 1093 | root_blk->bp->b_ops = bp->b_ops; |
| 1094 | xfs_trans_buf_copy_type(root_blk->bp, bp); |
| 1095 | if (oldroothdr.magic == XFS_DA3_NODE_MAGIC) { |
| 1096 | struct xfs_da3_blkinfo *da3 = root_blk->bp->b_addr; |
| 1097 | da3->blkno = cpu_to_be64(root_blk->bp->b_bn); |
| 1098 | } |
| 1099 | xfs_trans_log_buf(args->trans, root_blk->bp, 0, state->blocksize - 1); |
| 1100 | error = xfs_da_shrink_inode(args, child, bp); |
| 1101 | return(error); |
| 1102 | } |
| 1103 | |
| 1104 | /* |
| 1105 | * Check a node block and its neighbors to see if the block should be |
| 1106 | * collapsed into one or the other neighbor. Always keep the block |
| 1107 | * with the smaller block number. |
| 1108 | * If the current block is over 50% full, don't try to join it, return 0. |
| 1109 | * If the block is empty, fill in the state structure and return 2. |
| 1110 | * If it can be collapsed, fill in the state structure and return 1. |
| 1111 | * If nothing can be done, return 0. |
| 1112 | */ |
| 1113 | STATIC int |
| 1114 | xfs_da3_node_toosmall( |
| 1115 | struct xfs_da_state *state, |
| 1116 | int *action) |
| 1117 | { |
| 1118 | struct xfs_da_intnode *node; |
| 1119 | struct xfs_da_state_blk *blk; |
| 1120 | struct xfs_da_blkinfo *info; |
| 1121 | xfs_dablk_t blkno; |
| 1122 | struct xfs_buf *bp; |
| 1123 | struct xfs_da3_icnode_hdr nodehdr; |
| 1124 | int count; |
| 1125 | int forward; |
| 1126 | int error; |
| 1127 | int retval; |
| 1128 | int i; |
| 1129 | struct xfs_inode *dp = state->args->dp; |
| 1130 | |
| 1131 | trace_xfs_da_node_toosmall(state->args); |
| 1132 | |
| 1133 | /* |
| 1134 | * Check for the degenerate case of the block being over 50% full. |
| 1135 | * If so, it's not worth even looking to see if we might be able |
| 1136 | * to coalesce with a sibling. |
| 1137 | */ |
| 1138 | blk = &state->path.blk[ state->path.active-1 ]; |
| 1139 | info = blk->bp->b_addr; |
| 1140 | node = (xfs_da_intnode_t *)info; |
| 1141 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
| 1142 | if (nodehdr.count > (state->node_ents >> 1)) { |
| 1143 | *action = 0; /* blk over 50%, don't try to join */ |
| 1144 | return(0); /* blk over 50%, don't try to join */ |
| 1145 | } |
| 1146 | |
| 1147 | /* |
| 1148 | * Check for the degenerate case of the block being empty. |
| 1149 | * If the block is empty, we'll simply delete it, no need to |
| 1150 | * coalesce it with a sibling block. We choose (arbitrarily) |
| 1151 | * to merge with the forward block unless it is NULL. |
| 1152 | */ |
| 1153 | if (nodehdr.count == 0) { |
| 1154 | /* |
| 1155 | * Make altpath point to the block we want to keep and |
| 1156 | * path point to the block we want to drop (this one). |
| 1157 | */ |
| 1158 | forward = (info->forw != 0); |
| 1159 | memcpy(&state->altpath, &state->path, sizeof(state->path)); |
| 1160 | error = xfs_da3_path_shift(state, &state->altpath, forward, |
| 1161 | 0, &retval); |
| 1162 | if (error) |
| 1163 | return(error); |
| 1164 | if (retval) { |
| 1165 | *action = 0; |
| 1166 | } else { |
| 1167 | *action = 2; |
| 1168 | } |
| 1169 | return(0); |
| 1170 | } |
| 1171 | |
| 1172 | /* |
| 1173 | * Examine each sibling block to see if we can coalesce with |
| 1174 | * at least 25% free space to spare. We need to figure out |
| 1175 | * whether to merge with the forward or the backward block. |
| 1176 | * We prefer coalescing with the lower numbered sibling so as |
| 1177 | * to shrink a directory over time. |
| 1178 | */ |
| 1179 | count = state->node_ents; |
| 1180 | count -= state->node_ents >> 2; |
| 1181 | count -= nodehdr.count; |
| 1182 | |
| 1183 | /* start with smaller blk num */ |
| 1184 | forward = nodehdr.forw < nodehdr.back; |
| 1185 | for (i = 0; i < 2; forward = !forward, i++) { |
| 1186 | struct xfs_da3_icnode_hdr thdr; |
| 1187 | if (forward) |
| 1188 | blkno = nodehdr.forw; |
| 1189 | else |
| 1190 | blkno = nodehdr.back; |
| 1191 | if (blkno == 0) |
| 1192 | continue; |
| 1193 | error = xfs_da3_node_read(state->args->trans, dp, |
| 1194 | blkno, -1, &bp, state->args->whichfork); |
| 1195 | if (error) |
| 1196 | return(error); |
| 1197 | |
| 1198 | node = bp->b_addr; |
| 1199 | dp->d_ops->node_hdr_from_disk(&thdr, node); |
| 1200 | xfs_trans_brelse(state->args->trans, bp); |
| 1201 | |
| 1202 | if (count - thdr.count >= 0) |
| 1203 | break; /* fits with at least 25% to spare */ |
| 1204 | } |
| 1205 | if (i >= 2) { |
| 1206 | *action = 0; |
| 1207 | return 0; |
| 1208 | } |
| 1209 | |
| 1210 | /* |
| 1211 | * Make altpath point to the block we want to keep (the lower |
| 1212 | * numbered block) and path point to the block we want to drop. |
| 1213 | */ |
| 1214 | memcpy(&state->altpath, &state->path, sizeof(state->path)); |
| 1215 | if (blkno < blk->blkno) { |
| 1216 | error = xfs_da3_path_shift(state, &state->altpath, forward, |
| 1217 | 0, &retval); |
| 1218 | } else { |
| 1219 | error = xfs_da3_path_shift(state, &state->path, forward, |
| 1220 | 0, &retval); |
| 1221 | } |
| 1222 | if (error) |
| 1223 | return error; |
| 1224 | if (retval) { |
| 1225 | *action = 0; |
| 1226 | return 0; |
| 1227 | } |
| 1228 | *action = 1; |
| 1229 | return 0; |
| 1230 | } |
| 1231 | |
| 1232 | /* |
| 1233 | * Pick up the last hashvalue from an intermediate node. |
| 1234 | */ |
| 1235 | STATIC uint |
| 1236 | xfs_da3_node_lasthash( |
| 1237 | struct xfs_inode *dp, |
| 1238 | struct xfs_buf *bp, |
| 1239 | int *count) |
| 1240 | { |
| 1241 | struct xfs_da_intnode *node; |
| 1242 | struct xfs_da_node_entry *btree; |
| 1243 | struct xfs_da3_icnode_hdr nodehdr; |
| 1244 | |
| 1245 | node = bp->b_addr; |
| 1246 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
| 1247 | if (count) |
| 1248 | *count = nodehdr.count; |
| 1249 | if (!nodehdr.count) |
| 1250 | return 0; |
| 1251 | btree = dp->d_ops->node_tree_p(node); |
| 1252 | return be32_to_cpu(btree[nodehdr.count - 1].hashval); |
| 1253 | } |
| 1254 | |
| 1255 | /* |
| 1256 | * Walk back up the tree adjusting hash values as necessary, |
| 1257 | * when we stop making changes, return. |
| 1258 | */ |
| 1259 | void |
| 1260 | xfs_da3_fixhashpath( |
| 1261 | struct xfs_da_state *state, |
| 1262 | struct xfs_da_state_path *path) |
| 1263 | { |
| 1264 | struct xfs_da_state_blk *blk; |
| 1265 | struct xfs_da_intnode *node; |
| 1266 | struct xfs_da_node_entry *btree; |
| 1267 | xfs_dahash_t lasthash=0; |
| 1268 | int level; |
| 1269 | int count; |
| 1270 | struct xfs_inode *dp = state->args->dp; |
| 1271 | |
| 1272 | trace_xfs_da_fixhashpath(state->args); |
| 1273 | |
| 1274 | level = path->active-1; |
| 1275 | blk = &path->blk[ level ]; |
| 1276 | switch (blk->magic) { |
| 1277 | case XFS_ATTR_LEAF_MAGIC: |
| 1278 | lasthash = xfs_attr_leaf_lasthash(blk->bp, &count); |
| 1279 | if (count == 0) |
| 1280 | return; |
| 1281 | break; |
| 1282 | case XFS_DIR2_LEAFN_MAGIC: |
| 1283 | lasthash = xfs_dir2_leafn_lasthash(dp, blk->bp, &count); |
| 1284 | if (count == 0) |
| 1285 | return; |
| 1286 | break; |
| 1287 | case XFS_DA_NODE_MAGIC: |
| 1288 | lasthash = xfs_da3_node_lasthash(dp, blk->bp, &count); |
| 1289 | if (count == 0) |
| 1290 | return; |
| 1291 | break; |
| 1292 | } |
| 1293 | for (blk--, level--; level >= 0; blk--, level--) { |
| 1294 | struct xfs_da3_icnode_hdr nodehdr; |
| 1295 | |
| 1296 | node = blk->bp->b_addr; |
| 1297 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
| 1298 | btree = dp->d_ops->node_tree_p(node); |
| 1299 | if (be32_to_cpu(btree[blk->index].hashval) == lasthash) |
| 1300 | break; |
| 1301 | blk->hashval = lasthash; |
| 1302 | btree[blk->index].hashval = cpu_to_be32(lasthash); |
| 1303 | xfs_trans_log_buf(state->args->trans, blk->bp, |
| 1304 | XFS_DA_LOGRANGE(node, &btree[blk->index], |
| 1305 | sizeof(*btree))); |
| 1306 | |
| 1307 | lasthash = be32_to_cpu(btree[nodehdr.count - 1].hashval); |
| 1308 | } |
| 1309 | } |
| 1310 | |
| 1311 | /* |
| 1312 | * Remove an entry from an intermediate node. |
| 1313 | */ |
| 1314 | STATIC void |
| 1315 | xfs_da3_node_remove( |
| 1316 | struct xfs_da_state *state, |
| 1317 | struct xfs_da_state_blk *drop_blk) |
| 1318 | { |
| 1319 | struct xfs_da_intnode *node; |
| 1320 | struct xfs_da3_icnode_hdr nodehdr; |
| 1321 | struct xfs_da_node_entry *btree; |
| 1322 | int index; |
| 1323 | int tmp; |
| 1324 | struct xfs_inode *dp = state->args->dp; |
| 1325 | |
| 1326 | trace_xfs_da_node_remove(state->args); |
| 1327 | |
| 1328 | node = drop_blk->bp->b_addr; |
| 1329 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
| 1330 | ASSERT(drop_blk->index < nodehdr.count); |
| 1331 | ASSERT(drop_blk->index >= 0); |
| 1332 | |
| 1333 | /* |
| 1334 | * Copy over the offending entry, or just zero it out. |
| 1335 | */ |
| 1336 | index = drop_blk->index; |
| 1337 | btree = dp->d_ops->node_tree_p(node); |
| 1338 | if (index < nodehdr.count - 1) { |
| 1339 | tmp = nodehdr.count - index - 1; |
| 1340 | tmp *= (uint)sizeof(xfs_da_node_entry_t); |
| 1341 | memmove(&btree[index], &btree[index + 1], tmp); |
| 1342 | xfs_trans_log_buf(state->args->trans, drop_blk->bp, |
| 1343 | XFS_DA_LOGRANGE(node, &btree[index], tmp)); |
| 1344 | index = nodehdr.count - 1; |
| 1345 | } |
| 1346 | memset(&btree[index], 0, sizeof(xfs_da_node_entry_t)); |
| 1347 | xfs_trans_log_buf(state->args->trans, drop_blk->bp, |
| 1348 | XFS_DA_LOGRANGE(node, &btree[index], sizeof(btree[index]))); |
| 1349 | nodehdr.count -= 1; |
| 1350 | dp->d_ops->node_hdr_to_disk(node, &nodehdr); |
| 1351 | xfs_trans_log_buf(state->args->trans, drop_blk->bp, |
| 1352 | XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size)); |
| 1353 | |
| 1354 | /* |
| 1355 | * Copy the last hash value from the block to propagate upwards. |
| 1356 | */ |
| 1357 | drop_blk->hashval = be32_to_cpu(btree[index - 1].hashval); |
| 1358 | } |
| 1359 | |
| 1360 | /* |
| 1361 | * Unbalance the elements between two intermediate nodes, |
| 1362 | * move all Btree elements from one node into another. |
| 1363 | */ |
| 1364 | STATIC void |
| 1365 | xfs_da3_node_unbalance( |
| 1366 | struct xfs_da_state *state, |
| 1367 | struct xfs_da_state_blk *drop_blk, |
| 1368 | struct xfs_da_state_blk *save_blk) |
| 1369 | { |
| 1370 | struct xfs_da_intnode *drop_node; |
| 1371 | struct xfs_da_intnode *save_node; |
| 1372 | struct xfs_da_node_entry *drop_btree; |
| 1373 | struct xfs_da_node_entry *save_btree; |
| 1374 | struct xfs_da3_icnode_hdr drop_hdr; |
| 1375 | struct xfs_da3_icnode_hdr save_hdr; |
| 1376 | struct xfs_trans *tp; |
| 1377 | int sindex; |
| 1378 | int tmp; |
| 1379 | struct xfs_inode *dp = state->args->dp; |
| 1380 | |
| 1381 | trace_xfs_da_node_unbalance(state->args); |
| 1382 | |
| 1383 | drop_node = drop_blk->bp->b_addr; |
| 1384 | save_node = save_blk->bp->b_addr; |
| 1385 | dp->d_ops->node_hdr_from_disk(&drop_hdr, drop_node); |
| 1386 | dp->d_ops->node_hdr_from_disk(&save_hdr, save_node); |
| 1387 | drop_btree = dp->d_ops->node_tree_p(drop_node); |
| 1388 | save_btree = dp->d_ops->node_tree_p(save_node); |
| 1389 | tp = state->args->trans; |
| 1390 | |
| 1391 | /* |
| 1392 | * If the dying block has lower hashvals, then move all the |
| 1393 | * elements in the remaining block up to make a hole. |
| 1394 | */ |
| 1395 | if ((be32_to_cpu(drop_btree[0].hashval) < |
| 1396 | be32_to_cpu(save_btree[0].hashval)) || |
| 1397 | (be32_to_cpu(drop_btree[drop_hdr.count - 1].hashval) < |
| 1398 | be32_to_cpu(save_btree[save_hdr.count - 1].hashval))) { |
| 1399 | /* XXX: check this - is memmove dst correct? */ |
| 1400 | tmp = save_hdr.count * sizeof(xfs_da_node_entry_t); |
| 1401 | memmove(&save_btree[drop_hdr.count], &save_btree[0], tmp); |
| 1402 | |
| 1403 | sindex = 0; |
| 1404 | xfs_trans_log_buf(tp, save_blk->bp, |
| 1405 | XFS_DA_LOGRANGE(save_node, &save_btree[0], |
| 1406 | (save_hdr.count + drop_hdr.count) * |
| 1407 | sizeof(xfs_da_node_entry_t))); |
| 1408 | } else { |
| 1409 | sindex = save_hdr.count; |
| 1410 | xfs_trans_log_buf(tp, save_blk->bp, |
| 1411 | XFS_DA_LOGRANGE(save_node, &save_btree[sindex], |
| 1412 | drop_hdr.count * sizeof(xfs_da_node_entry_t))); |
| 1413 | } |
| 1414 | |
| 1415 | /* |
| 1416 | * Move all the B-tree elements from drop_blk to save_blk. |
| 1417 | */ |
| 1418 | tmp = drop_hdr.count * (uint)sizeof(xfs_da_node_entry_t); |
| 1419 | memcpy(&save_btree[sindex], &drop_btree[0], tmp); |
| 1420 | save_hdr.count += drop_hdr.count; |
| 1421 | |
| 1422 | dp->d_ops->node_hdr_to_disk(save_node, &save_hdr); |
| 1423 | xfs_trans_log_buf(tp, save_blk->bp, |
| 1424 | XFS_DA_LOGRANGE(save_node, &save_node->hdr, |
| 1425 | dp->d_ops->node_hdr_size)); |
| 1426 | |
| 1427 | /* |
| 1428 | * Save the last hashval in the remaining block for upward propagation. |
| 1429 | */ |
| 1430 | save_blk->hashval = be32_to_cpu(save_btree[save_hdr.count - 1].hashval); |
| 1431 | } |
| 1432 | |
| 1433 | /*======================================================================== |
| 1434 | * Routines used for finding things in the Btree. |
| 1435 | *========================================================================*/ |
| 1436 | |
| 1437 | /* |
| 1438 | * Walk down the Btree looking for a particular filename, filling |
| 1439 | * in the state structure as we go. |
| 1440 | * |
| 1441 | * We will set the state structure to point to each of the elements |
| 1442 | * in each of the nodes where either the hashval is or should be. |
| 1443 | * |
| 1444 | * We support duplicate hashval's so for each entry in the current |
| 1445 | * node that could contain the desired hashval, descend. This is a |
| 1446 | * pruned depth-first tree search. |
| 1447 | */ |
| 1448 | int /* error */ |
| 1449 | xfs_da3_node_lookup_int( |
| 1450 | struct xfs_da_state *state, |
| 1451 | int *result) |
| 1452 | { |
| 1453 | struct xfs_da_state_blk *blk; |
| 1454 | struct xfs_da_blkinfo *curr; |
| 1455 | struct xfs_da_intnode *node; |
| 1456 | struct xfs_da_node_entry *btree; |
| 1457 | struct xfs_da3_icnode_hdr nodehdr; |
| 1458 | struct xfs_da_args *args; |
| 1459 | xfs_dablk_t blkno; |
| 1460 | xfs_dahash_t hashval; |
| 1461 | xfs_dahash_t btreehashval; |
| 1462 | int probe; |
| 1463 | int span; |
| 1464 | int max; |
| 1465 | int error; |
| 1466 | int retval; |
| 1467 | struct xfs_inode *dp = state->args->dp; |
| 1468 | |
| 1469 | args = state->args; |
| 1470 | |
| 1471 | /* |
| 1472 | * Descend thru the B-tree searching each level for the right |
| 1473 | * node to use, until the right hashval is found. |
| 1474 | */ |
| 1475 | blkno = (args->whichfork == XFS_DATA_FORK)? args->geo->leafblk : 0; |
| 1476 | for (blk = &state->path.blk[0], state->path.active = 1; |
| 1477 | state->path.active <= XFS_DA_NODE_MAXDEPTH; |
| 1478 | blk++, state->path.active++) { |
| 1479 | /* |
| 1480 | * Read the next node down in the tree. |
| 1481 | */ |
| 1482 | blk->blkno = blkno; |
| 1483 | error = xfs_da3_node_read(args->trans, args->dp, blkno, |
| 1484 | -1, &blk->bp, args->whichfork); |
| 1485 | if (error) { |
| 1486 | blk->blkno = 0; |
| 1487 | state->path.active--; |
| 1488 | return(error); |
| 1489 | } |
| 1490 | curr = blk->bp->b_addr; |
| 1491 | blk->magic = be16_to_cpu(curr->magic); |
| 1492 | |
| 1493 | if (blk->magic == XFS_ATTR_LEAF_MAGIC || |
| 1494 | blk->magic == XFS_ATTR3_LEAF_MAGIC) { |
| 1495 | blk->magic = XFS_ATTR_LEAF_MAGIC; |
| 1496 | blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL); |
| 1497 | break; |
| 1498 | } |
| 1499 | |
| 1500 | if (blk->magic == XFS_DIR2_LEAFN_MAGIC || |
| 1501 | blk->magic == XFS_DIR3_LEAFN_MAGIC) { |
| 1502 | blk->magic = XFS_DIR2_LEAFN_MAGIC; |
| 1503 | blk->hashval = xfs_dir2_leafn_lasthash(args->dp, |
| 1504 | blk->bp, NULL); |
| 1505 | break; |
| 1506 | } |
| 1507 | |
| 1508 | blk->magic = XFS_DA_NODE_MAGIC; |
| 1509 | |
| 1510 | |
| 1511 | /* |
| 1512 | * Search an intermediate node for a match. |
| 1513 | */ |
| 1514 | node = blk->bp->b_addr; |
| 1515 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
| 1516 | btree = dp->d_ops->node_tree_p(node); |
| 1517 | |
| 1518 | max = nodehdr.count; |
| 1519 | blk->hashval = be32_to_cpu(btree[max - 1].hashval); |
| 1520 | |
| 1521 | /* |
| 1522 | * Binary search. (note: small blocks will skip loop) |
| 1523 | */ |
| 1524 | probe = span = max / 2; |
| 1525 | hashval = args->hashval; |
| 1526 | while (span > 4) { |
| 1527 | span /= 2; |
| 1528 | btreehashval = be32_to_cpu(btree[probe].hashval); |
| 1529 | if (btreehashval < hashval) |
| 1530 | probe += span; |
| 1531 | else if (btreehashval > hashval) |
| 1532 | probe -= span; |
| 1533 | else |
| 1534 | break; |
| 1535 | } |
| 1536 | ASSERT((probe >= 0) && (probe < max)); |
| 1537 | ASSERT((span <= 4) || |
| 1538 | (be32_to_cpu(btree[probe].hashval) == hashval)); |
| 1539 | |
| 1540 | /* |
| 1541 | * Since we may have duplicate hashval's, find the first |
| 1542 | * matching hashval in the node. |
| 1543 | */ |
| 1544 | while (probe > 0 && |
| 1545 | be32_to_cpu(btree[probe].hashval) >= hashval) { |
| 1546 | probe--; |
| 1547 | } |
| 1548 | while (probe < max && |
| 1549 | be32_to_cpu(btree[probe].hashval) < hashval) { |
| 1550 | probe++; |
| 1551 | } |
| 1552 | |
| 1553 | /* |
| 1554 | * Pick the right block to descend on. |
| 1555 | */ |
| 1556 | if (probe == max) { |
| 1557 | blk->index = max - 1; |
| 1558 | blkno = be32_to_cpu(btree[max - 1].before); |
| 1559 | } else { |
| 1560 | blk->index = probe; |
| 1561 | blkno = be32_to_cpu(btree[probe].before); |
| 1562 | } |
| 1563 | } |
| 1564 | |
| 1565 | /* |
| 1566 | * A leaf block that ends in the hashval that we are interested in |
| 1567 | * (final hashval == search hashval) means that the next block may |
| 1568 | * contain more entries with the same hashval, shift upward to the |
| 1569 | * next leaf and keep searching. |
| 1570 | */ |
| 1571 | for (;;) { |
| 1572 | if (blk->magic == XFS_DIR2_LEAFN_MAGIC) { |
| 1573 | retval = xfs_dir2_leafn_lookup_int(blk->bp, args, |
| 1574 | &blk->index, state); |
| 1575 | } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) { |
| 1576 | retval = xfs_attr3_leaf_lookup_int(blk->bp, args); |
| 1577 | blk->index = args->index; |
| 1578 | args->blkno = blk->blkno; |
| 1579 | } else { |
| 1580 | ASSERT(0); |
| 1581 | return XFS_ERROR(EFSCORRUPTED); |
| 1582 | } |
| 1583 | if (((retval == ENOENT) || (retval == ENOATTR)) && |
| 1584 | (blk->hashval == args->hashval)) { |
| 1585 | error = xfs_da3_path_shift(state, &state->path, 1, 1, |
| 1586 | &retval); |
| 1587 | if (error) |
| 1588 | return(error); |
| 1589 | if (retval == 0) { |
| 1590 | continue; |
| 1591 | } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) { |
| 1592 | /* path_shift() gives ENOENT */ |
| 1593 | retval = XFS_ERROR(ENOATTR); |
| 1594 | } |
| 1595 | } |
| 1596 | break; |
| 1597 | } |
| 1598 | *result = retval; |
| 1599 | return(0); |
| 1600 | } |
| 1601 | |
| 1602 | /*======================================================================== |
| 1603 | * Utility routines. |
| 1604 | *========================================================================*/ |
| 1605 | |
| 1606 | /* |
| 1607 | * Compare two intermediate nodes for "order". |
| 1608 | */ |
| 1609 | STATIC int |
| 1610 | xfs_da3_node_order( |
| 1611 | struct xfs_inode *dp, |
| 1612 | struct xfs_buf *node1_bp, |
| 1613 | struct xfs_buf *node2_bp) |
| 1614 | { |
| 1615 | struct xfs_da_intnode *node1; |
| 1616 | struct xfs_da_intnode *node2; |
| 1617 | struct xfs_da_node_entry *btree1; |
| 1618 | struct xfs_da_node_entry *btree2; |
| 1619 | struct xfs_da3_icnode_hdr node1hdr; |
| 1620 | struct xfs_da3_icnode_hdr node2hdr; |
| 1621 | |
| 1622 | node1 = node1_bp->b_addr; |
| 1623 | node2 = node2_bp->b_addr; |
| 1624 | dp->d_ops->node_hdr_from_disk(&node1hdr, node1); |
| 1625 | dp->d_ops->node_hdr_from_disk(&node2hdr, node2); |
| 1626 | btree1 = dp->d_ops->node_tree_p(node1); |
| 1627 | btree2 = dp->d_ops->node_tree_p(node2); |
| 1628 | |
| 1629 | if (node1hdr.count > 0 && node2hdr.count > 0 && |
| 1630 | ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) || |
| 1631 | (be32_to_cpu(btree2[node2hdr.count - 1].hashval) < |
| 1632 | be32_to_cpu(btree1[node1hdr.count - 1].hashval)))) { |
| 1633 | return 1; |
| 1634 | } |
| 1635 | return 0; |
| 1636 | } |
| 1637 | |
| 1638 | /* |
| 1639 | * Link a new block into a doubly linked list of blocks (of whatever type). |
| 1640 | */ |
| 1641 | int /* error */ |
| 1642 | xfs_da3_blk_link( |
| 1643 | struct xfs_da_state *state, |
| 1644 | struct xfs_da_state_blk *old_blk, |
| 1645 | struct xfs_da_state_blk *new_blk) |
| 1646 | { |
| 1647 | struct xfs_da_blkinfo *old_info; |
| 1648 | struct xfs_da_blkinfo *new_info; |
| 1649 | struct xfs_da_blkinfo *tmp_info; |
| 1650 | struct xfs_da_args *args; |
| 1651 | struct xfs_buf *bp; |
| 1652 | int before = 0; |
| 1653 | int error; |
| 1654 | struct xfs_inode *dp = state->args->dp; |
| 1655 | |
| 1656 | /* |
| 1657 | * Set up environment. |
| 1658 | */ |
| 1659 | args = state->args; |
| 1660 | ASSERT(args != NULL); |
| 1661 | old_info = old_blk->bp->b_addr; |
| 1662 | new_info = new_blk->bp->b_addr; |
| 1663 | ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC || |
| 1664 | old_blk->magic == XFS_DIR2_LEAFN_MAGIC || |
| 1665 | old_blk->magic == XFS_ATTR_LEAF_MAGIC); |
| 1666 | |
| 1667 | switch (old_blk->magic) { |
| 1668 | case XFS_ATTR_LEAF_MAGIC: |
| 1669 | before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp); |
| 1670 | break; |
| 1671 | case XFS_DIR2_LEAFN_MAGIC: |
| 1672 | before = xfs_dir2_leafn_order(dp, old_blk->bp, new_blk->bp); |
| 1673 | break; |
| 1674 | case XFS_DA_NODE_MAGIC: |
| 1675 | before = xfs_da3_node_order(dp, old_blk->bp, new_blk->bp); |
| 1676 | break; |
| 1677 | } |
| 1678 | |
| 1679 | /* |
| 1680 | * Link blocks in appropriate order. |
| 1681 | */ |
| 1682 | if (before) { |
| 1683 | /* |
| 1684 | * Link new block in before existing block. |
| 1685 | */ |
| 1686 | trace_xfs_da_link_before(args); |
| 1687 | new_info->forw = cpu_to_be32(old_blk->blkno); |
| 1688 | new_info->back = old_info->back; |
| 1689 | if (old_info->back) { |
| 1690 | error = xfs_da3_node_read(args->trans, dp, |
| 1691 | be32_to_cpu(old_info->back), |
| 1692 | -1, &bp, args->whichfork); |
| 1693 | if (error) |
| 1694 | return(error); |
| 1695 | ASSERT(bp != NULL); |
| 1696 | tmp_info = bp->b_addr; |
| 1697 | ASSERT(tmp_info->magic == old_info->magic); |
| 1698 | ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno); |
| 1699 | tmp_info->forw = cpu_to_be32(new_blk->blkno); |
| 1700 | xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1); |
| 1701 | } |
| 1702 | old_info->back = cpu_to_be32(new_blk->blkno); |
| 1703 | } else { |
| 1704 | /* |
| 1705 | * Link new block in after existing block. |
| 1706 | */ |
| 1707 | trace_xfs_da_link_after(args); |
| 1708 | new_info->forw = old_info->forw; |
| 1709 | new_info->back = cpu_to_be32(old_blk->blkno); |
| 1710 | if (old_info->forw) { |
| 1711 | error = xfs_da3_node_read(args->trans, dp, |
| 1712 | be32_to_cpu(old_info->forw), |
| 1713 | -1, &bp, args->whichfork); |
| 1714 | if (error) |
| 1715 | return(error); |
| 1716 | ASSERT(bp != NULL); |
| 1717 | tmp_info = bp->b_addr; |
| 1718 | ASSERT(tmp_info->magic == old_info->magic); |
| 1719 | ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno); |
| 1720 | tmp_info->back = cpu_to_be32(new_blk->blkno); |
| 1721 | xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1); |
| 1722 | } |
| 1723 | old_info->forw = cpu_to_be32(new_blk->blkno); |
| 1724 | } |
| 1725 | |
| 1726 | xfs_trans_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1); |
| 1727 | xfs_trans_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1); |
| 1728 | return(0); |
| 1729 | } |
| 1730 | |
| 1731 | /* |
| 1732 | * Unlink a block from a doubly linked list of blocks. |
| 1733 | */ |
| 1734 | STATIC int /* error */ |
| 1735 | xfs_da3_blk_unlink( |
| 1736 | struct xfs_da_state *state, |
| 1737 | struct xfs_da_state_blk *drop_blk, |
| 1738 | struct xfs_da_state_blk *save_blk) |
| 1739 | { |
| 1740 | struct xfs_da_blkinfo *drop_info; |
| 1741 | struct xfs_da_blkinfo *save_info; |
| 1742 | struct xfs_da_blkinfo *tmp_info; |
| 1743 | struct xfs_da_args *args; |
| 1744 | struct xfs_buf *bp; |
| 1745 | int error; |
| 1746 | |
| 1747 | /* |
| 1748 | * Set up environment. |
| 1749 | */ |
| 1750 | args = state->args; |
| 1751 | ASSERT(args != NULL); |
| 1752 | save_info = save_blk->bp->b_addr; |
| 1753 | drop_info = drop_blk->bp->b_addr; |
| 1754 | ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC || |
| 1755 | save_blk->magic == XFS_DIR2_LEAFN_MAGIC || |
| 1756 | save_blk->magic == XFS_ATTR_LEAF_MAGIC); |
| 1757 | ASSERT(save_blk->magic == drop_blk->magic); |
| 1758 | ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) || |
| 1759 | (be32_to_cpu(save_info->back) == drop_blk->blkno)); |
| 1760 | ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) || |
| 1761 | (be32_to_cpu(drop_info->back) == save_blk->blkno)); |
| 1762 | |
| 1763 | /* |
| 1764 | * Unlink the leaf block from the doubly linked chain of leaves. |
| 1765 | */ |
| 1766 | if (be32_to_cpu(save_info->back) == drop_blk->blkno) { |
| 1767 | trace_xfs_da_unlink_back(args); |
| 1768 | save_info->back = drop_info->back; |
| 1769 | if (drop_info->back) { |
| 1770 | error = xfs_da3_node_read(args->trans, args->dp, |
| 1771 | be32_to_cpu(drop_info->back), |
| 1772 | -1, &bp, args->whichfork); |
| 1773 | if (error) |
| 1774 | return(error); |
| 1775 | ASSERT(bp != NULL); |
| 1776 | tmp_info = bp->b_addr; |
| 1777 | ASSERT(tmp_info->magic == save_info->magic); |
| 1778 | ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno); |
| 1779 | tmp_info->forw = cpu_to_be32(save_blk->blkno); |
| 1780 | xfs_trans_log_buf(args->trans, bp, 0, |
| 1781 | sizeof(*tmp_info) - 1); |
| 1782 | } |
| 1783 | } else { |
| 1784 | trace_xfs_da_unlink_forward(args); |
| 1785 | save_info->forw = drop_info->forw; |
| 1786 | if (drop_info->forw) { |
| 1787 | error = xfs_da3_node_read(args->trans, args->dp, |
| 1788 | be32_to_cpu(drop_info->forw), |
| 1789 | -1, &bp, args->whichfork); |
| 1790 | if (error) |
| 1791 | return(error); |
| 1792 | ASSERT(bp != NULL); |
| 1793 | tmp_info = bp->b_addr; |
| 1794 | ASSERT(tmp_info->magic == save_info->magic); |
| 1795 | ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno); |
| 1796 | tmp_info->back = cpu_to_be32(save_blk->blkno); |
| 1797 | xfs_trans_log_buf(args->trans, bp, 0, |
| 1798 | sizeof(*tmp_info) - 1); |
| 1799 | } |
| 1800 | } |
| 1801 | |
| 1802 | xfs_trans_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1); |
| 1803 | return(0); |
| 1804 | } |
| 1805 | |
| 1806 | /* |
| 1807 | * Move a path "forward" or "!forward" one block at the current level. |
| 1808 | * |
| 1809 | * This routine will adjust a "path" to point to the next block |
| 1810 | * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the |
| 1811 | * Btree, including updating pointers to the intermediate nodes between |
| 1812 | * the new bottom and the root. |
| 1813 | */ |
| 1814 | int /* error */ |
| 1815 | xfs_da3_path_shift( |
| 1816 | struct xfs_da_state *state, |
| 1817 | struct xfs_da_state_path *path, |
| 1818 | int forward, |
| 1819 | int release, |
| 1820 | int *result) |
| 1821 | { |
| 1822 | struct xfs_da_state_blk *blk; |
| 1823 | struct xfs_da_blkinfo *info; |
| 1824 | struct xfs_da_intnode *node; |
| 1825 | struct xfs_da_args *args; |
| 1826 | struct xfs_da_node_entry *btree; |
| 1827 | struct xfs_da3_icnode_hdr nodehdr; |
| 1828 | xfs_dablk_t blkno = 0; |
| 1829 | int level; |
| 1830 | int error; |
| 1831 | struct xfs_inode *dp = state->args->dp; |
| 1832 | |
| 1833 | trace_xfs_da_path_shift(state->args); |
| 1834 | |
| 1835 | /* |
| 1836 | * Roll up the Btree looking for the first block where our |
| 1837 | * current index is not at the edge of the block. Note that |
| 1838 | * we skip the bottom layer because we want the sibling block. |
| 1839 | */ |
| 1840 | args = state->args; |
| 1841 | ASSERT(args != NULL); |
| 1842 | ASSERT(path != NULL); |
| 1843 | ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH)); |
| 1844 | level = (path->active-1) - 1; /* skip bottom layer in path */ |
| 1845 | for (blk = &path->blk[level]; level >= 0; blk--, level--) { |
| 1846 | node = blk->bp->b_addr; |
| 1847 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
| 1848 | btree = dp->d_ops->node_tree_p(node); |
| 1849 | |
| 1850 | if (forward && (blk->index < nodehdr.count - 1)) { |
| 1851 | blk->index++; |
| 1852 | blkno = be32_to_cpu(btree[blk->index].before); |
| 1853 | break; |
| 1854 | } else if (!forward && (blk->index > 0)) { |
| 1855 | blk->index--; |
| 1856 | blkno = be32_to_cpu(btree[blk->index].before); |
| 1857 | break; |
| 1858 | } |
| 1859 | } |
| 1860 | if (level < 0) { |
| 1861 | *result = XFS_ERROR(ENOENT); /* we're out of our tree */ |
| 1862 | ASSERT(args->op_flags & XFS_DA_OP_OKNOENT); |
| 1863 | return(0); |
| 1864 | } |
| 1865 | |
| 1866 | /* |
| 1867 | * Roll down the edge of the subtree until we reach the |
| 1868 | * same depth we were at originally. |
| 1869 | */ |
| 1870 | for (blk++, level++; level < path->active; blk++, level++) { |
| 1871 | /* |
| 1872 | * Release the old block. |
| 1873 | * (if it's dirty, trans won't actually let go) |
| 1874 | */ |
| 1875 | if (release) |
| 1876 | xfs_trans_brelse(args->trans, blk->bp); |
| 1877 | |
| 1878 | /* |
| 1879 | * Read the next child block. |
| 1880 | */ |
| 1881 | blk->blkno = blkno; |
| 1882 | error = xfs_da3_node_read(args->trans, dp, blkno, -1, |
| 1883 | &blk->bp, args->whichfork); |
| 1884 | if (error) |
| 1885 | return(error); |
| 1886 | info = blk->bp->b_addr; |
| 1887 | ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC) || |
| 1888 | info->magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) || |
| 1889 | info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) || |
| 1890 | info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) || |
| 1891 | info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) || |
| 1892 | info->magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)); |
| 1893 | |
| 1894 | |
| 1895 | /* |
| 1896 | * Note: we flatten the magic number to a single type so we |
| 1897 | * don't have to compare against crc/non-crc types elsewhere. |
| 1898 | */ |
| 1899 | switch (be16_to_cpu(info->magic)) { |
| 1900 | case XFS_DA_NODE_MAGIC: |
| 1901 | case XFS_DA3_NODE_MAGIC: |
| 1902 | blk->magic = XFS_DA_NODE_MAGIC; |
| 1903 | node = (xfs_da_intnode_t *)info; |
| 1904 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
| 1905 | btree = dp->d_ops->node_tree_p(node); |
| 1906 | blk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval); |
| 1907 | if (forward) |
| 1908 | blk->index = 0; |
| 1909 | else |
| 1910 | blk->index = nodehdr.count - 1; |
| 1911 | blkno = be32_to_cpu(btree[blk->index].before); |
| 1912 | break; |
| 1913 | case XFS_ATTR_LEAF_MAGIC: |
| 1914 | case XFS_ATTR3_LEAF_MAGIC: |
| 1915 | blk->magic = XFS_ATTR_LEAF_MAGIC; |
| 1916 | ASSERT(level == path->active-1); |
| 1917 | blk->index = 0; |
| 1918 | blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL); |
| 1919 | break; |
| 1920 | case XFS_DIR2_LEAFN_MAGIC: |
| 1921 | case XFS_DIR3_LEAFN_MAGIC: |
| 1922 | blk->magic = XFS_DIR2_LEAFN_MAGIC; |
| 1923 | ASSERT(level == path->active-1); |
| 1924 | blk->index = 0; |
| 1925 | blk->hashval = xfs_dir2_leafn_lasthash(args->dp, |
| 1926 | blk->bp, NULL); |
| 1927 | break; |
| 1928 | default: |
| 1929 | ASSERT(0); |
| 1930 | break; |
| 1931 | } |
| 1932 | } |
| 1933 | *result = 0; |
| 1934 | return 0; |
| 1935 | } |
| 1936 | |
| 1937 | |
| 1938 | /*======================================================================== |
| 1939 | * Utility routines. |
| 1940 | *========================================================================*/ |
| 1941 | |
| 1942 | /* |
| 1943 | * Implement a simple hash on a character string. |
| 1944 | * Rotate the hash value by 7 bits, then XOR each character in. |
| 1945 | * This is implemented with some source-level loop unrolling. |
| 1946 | */ |
| 1947 | xfs_dahash_t |
| 1948 | xfs_da_hashname(const __uint8_t *name, int namelen) |
| 1949 | { |
| 1950 | xfs_dahash_t hash; |
| 1951 | |
| 1952 | /* |
| 1953 | * Do four characters at a time as long as we can. |
| 1954 | */ |
| 1955 | for (hash = 0; namelen >= 4; namelen -= 4, name += 4) |
| 1956 | hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^ |
| 1957 | (name[3] << 0) ^ rol32(hash, 7 * 4); |
| 1958 | |
| 1959 | /* |
| 1960 | * Now do the rest of the characters. |
| 1961 | */ |
| 1962 | switch (namelen) { |
| 1963 | case 3: |
| 1964 | return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^ |
| 1965 | rol32(hash, 7 * 3); |
| 1966 | case 2: |
| 1967 | return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2); |
| 1968 | case 1: |
| 1969 | return (name[0] << 0) ^ rol32(hash, 7 * 1); |
| 1970 | default: /* case 0: */ |
| 1971 | return hash; |
| 1972 | } |
| 1973 | } |
| 1974 | |
| 1975 | enum xfs_dacmp |
| 1976 | xfs_da_compname( |
| 1977 | struct xfs_da_args *args, |
| 1978 | const unsigned char *name, |
| 1979 | int len) |
| 1980 | { |
| 1981 | return (args->namelen == len && memcmp(args->name, name, len) == 0) ? |
| 1982 | XFS_CMP_EXACT : XFS_CMP_DIFFERENT; |
| 1983 | } |
| 1984 | |
| 1985 | static xfs_dahash_t |
| 1986 | xfs_default_hashname( |
| 1987 | struct xfs_name *name) |
| 1988 | { |
| 1989 | return xfs_da_hashname(name->name, name->len); |
| 1990 | } |
| 1991 | |
| 1992 | const struct xfs_nameops xfs_default_nameops = { |
| 1993 | .hashname = xfs_default_hashname, |
| 1994 | .compname = xfs_da_compname |
| 1995 | }; |
| 1996 | |
| 1997 | int |
| 1998 | xfs_da_grow_inode_int( |
| 1999 | struct xfs_da_args *args, |
| 2000 | xfs_fileoff_t *bno, |
| 2001 | int count) |
| 2002 | { |
| 2003 | struct xfs_trans *tp = args->trans; |
| 2004 | struct xfs_inode *dp = args->dp; |
| 2005 | int w = args->whichfork; |
| 2006 | xfs_drfsbno_t nblks = dp->i_d.di_nblocks; |
| 2007 | struct xfs_bmbt_irec map, *mapp; |
| 2008 | int nmap, error, got, i, mapi; |
| 2009 | |
| 2010 | /* |
| 2011 | * Find a spot in the file space to put the new block. |
| 2012 | */ |
| 2013 | error = xfs_bmap_first_unused(tp, dp, count, bno, w); |
| 2014 | if (error) |
| 2015 | return error; |
| 2016 | |
| 2017 | /* |
| 2018 | * Try mapping it in one filesystem block. |
| 2019 | */ |
| 2020 | nmap = 1; |
| 2021 | ASSERT(args->firstblock != NULL); |
| 2022 | error = xfs_bmapi_write(tp, dp, *bno, count, |
| 2023 | xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA|XFS_BMAPI_CONTIG, |
| 2024 | args->firstblock, args->total, &map, &nmap, |
| 2025 | args->flist); |
| 2026 | if (error) |
| 2027 | return error; |
| 2028 | |
| 2029 | ASSERT(nmap <= 1); |
| 2030 | if (nmap == 1) { |
| 2031 | mapp = ↦ |
| 2032 | mapi = 1; |
| 2033 | } else if (nmap == 0 && count > 1) { |
| 2034 | xfs_fileoff_t b; |
| 2035 | int c; |
| 2036 | |
| 2037 | /* |
| 2038 | * If we didn't get it and the block might work if fragmented, |
| 2039 | * try without the CONTIG flag. Loop until we get it all. |
| 2040 | */ |
| 2041 | mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP); |
| 2042 | for (b = *bno, mapi = 0; b < *bno + count; ) { |
| 2043 | nmap = MIN(XFS_BMAP_MAX_NMAP, count); |
| 2044 | c = (int)(*bno + count - b); |
| 2045 | error = xfs_bmapi_write(tp, dp, b, c, |
| 2046 | xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA, |
| 2047 | args->firstblock, args->total, |
| 2048 | &mapp[mapi], &nmap, args->flist); |
| 2049 | if (error) |
| 2050 | goto out_free_map; |
| 2051 | if (nmap < 1) |
| 2052 | break; |
| 2053 | mapi += nmap; |
| 2054 | b = mapp[mapi - 1].br_startoff + |
| 2055 | mapp[mapi - 1].br_blockcount; |
| 2056 | } |
| 2057 | } else { |
| 2058 | mapi = 0; |
| 2059 | mapp = NULL; |
| 2060 | } |
| 2061 | |
| 2062 | /* |
| 2063 | * Count the blocks we got, make sure it matches the total. |
| 2064 | */ |
| 2065 | for (i = 0, got = 0; i < mapi; i++) |
| 2066 | got += mapp[i].br_blockcount; |
| 2067 | if (got != count || mapp[0].br_startoff != *bno || |
| 2068 | mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount != |
| 2069 | *bno + count) { |
| 2070 | error = XFS_ERROR(ENOSPC); |
| 2071 | goto out_free_map; |
| 2072 | } |
| 2073 | |
| 2074 | /* account for newly allocated blocks in reserved blocks total */ |
| 2075 | args->total -= dp->i_d.di_nblocks - nblks; |
| 2076 | |
| 2077 | out_free_map: |
| 2078 | if (mapp != &map) |
| 2079 | kmem_free(mapp); |
| 2080 | return error; |
| 2081 | } |
| 2082 | |
| 2083 | /* |
| 2084 | * Add a block to the btree ahead of the file. |
| 2085 | * Return the new block number to the caller. |
| 2086 | */ |
| 2087 | int |
| 2088 | xfs_da_grow_inode( |
| 2089 | struct xfs_da_args *args, |
| 2090 | xfs_dablk_t *new_blkno) |
| 2091 | { |
| 2092 | xfs_fileoff_t bno; |
| 2093 | int count; |
| 2094 | int error; |
| 2095 | |
| 2096 | trace_xfs_da_grow_inode(args); |
| 2097 | |
| 2098 | if (args->whichfork == XFS_DATA_FORK) { |
| 2099 | bno = args->geo->leafblk; |
| 2100 | count = args->dp->i_mount->m_dirblkfsbs; |
| 2101 | } else { |
| 2102 | bno = 0; |
| 2103 | count = 1; |
| 2104 | } |
| 2105 | |
| 2106 | error = xfs_da_grow_inode_int(args, &bno, count); |
| 2107 | if (!error) |
| 2108 | *new_blkno = (xfs_dablk_t)bno; |
| 2109 | return error; |
| 2110 | } |
| 2111 | |
| 2112 | /* |
| 2113 | * Ick. We need to always be able to remove a btree block, even |
| 2114 | * if there's no space reservation because the filesystem is full. |
| 2115 | * This is called if xfs_bunmapi on a btree block fails due to ENOSPC. |
| 2116 | * It swaps the target block with the last block in the file. The |
| 2117 | * last block in the file can always be removed since it can't cause |
| 2118 | * a bmap btree split to do that. |
| 2119 | */ |
| 2120 | STATIC int |
| 2121 | xfs_da3_swap_lastblock( |
| 2122 | struct xfs_da_args *args, |
| 2123 | xfs_dablk_t *dead_blknop, |
| 2124 | struct xfs_buf **dead_bufp) |
| 2125 | { |
| 2126 | struct xfs_da_blkinfo *dead_info; |
| 2127 | struct xfs_da_blkinfo *sib_info; |
| 2128 | struct xfs_da_intnode *par_node; |
| 2129 | struct xfs_da_intnode *dead_node; |
| 2130 | struct xfs_dir2_leaf *dead_leaf2; |
| 2131 | struct xfs_da_node_entry *btree; |
| 2132 | struct xfs_da3_icnode_hdr par_hdr; |
| 2133 | struct xfs_inode *dp; |
| 2134 | struct xfs_trans *tp; |
| 2135 | struct xfs_mount *mp; |
| 2136 | struct xfs_buf *dead_buf; |
| 2137 | struct xfs_buf *last_buf; |
| 2138 | struct xfs_buf *sib_buf; |
| 2139 | struct xfs_buf *par_buf; |
| 2140 | xfs_dahash_t dead_hash; |
| 2141 | xfs_fileoff_t lastoff; |
| 2142 | xfs_dablk_t dead_blkno; |
| 2143 | xfs_dablk_t last_blkno; |
| 2144 | xfs_dablk_t sib_blkno; |
| 2145 | xfs_dablk_t par_blkno; |
| 2146 | int error; |
| 2147 | int w; |
| 2148 | int entno; |
| 2149 | int level; |
| 2150 | int dead_level; |
| 2151 | |
| 2152 | trace_xfs_da_swap_lastblock(args); |
| 2153 | |
| 2154 | dead_buf = *dead_bufp; |
| 2155 | dead_blkno = *dead_blknop; |
| 2156 | tp = args->trans; |
| 2157 | dp = args->dp; |
| 2158 | w = args->whichfork; |
| 2159 | ASSERT(w == XFS_DATA_FORK); |
| 2160 | mp = dp->i_mount; |
| 2161 | lastoff = args->geo->freeblk; |
| 2162 | error = xfs_bmap_last_before(tp, dp, &lastoff, w); |
| 2163 | if (error) |
| 2164 | return error; |
| 2165 | if (unlikely(lastoff == 0)) { |
| 2166 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW, |
| 2167 | mp); |
| 2168 | return XFS_ERROR(EFSCORRUPTED); |
| 2169 | } |
| 2170 | /* |
| 2171 | * Read the last block in the btree space. |
| 2172 | */ |
| 2173 | last_blkno = (xfs_dablk_t)lastoff - mp->m_dirblkfsbs; |
| 2174 | error = xfs_da3_node_read(tp, dp, last_blkno, -1, &last_buf, w); |
| 2175 | if (error) |
| 2176 | return error; |
| 2177 | /* |
| 2178 | * Copy the last block into the dead buffer and log it. |
| 2179 | */ |
| 2180 | memcpy(dead_buf->b_addr, last_buf->b_addr, mp->m_dirblksize); |
| 2181 | xfs_trans_log_buf(tp, dead_buf, 0, mp->m_dirblksize - 1); |
| 2182 | dead_info = dead_buf->b_addr; |
| 2183 | /* |
| 2184 | * Get values from the moved block. |
| 2185 | */ |
| 2186 | if (dead_info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) || |
| 2187 | dead_info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) { |
| 2188 | struct xfs_dir3_icleaf_hdr leafhdr; |
| 2189 | struct xfs_dir2_leaf_entry *ents; |
| 2190 | |
| 2191 | dead_leaf2 = (xfs_dir2_leaf_t *)dead_info; |
| 2192 | dp->d_ops->leaf_hdr_from_disk(&leafhdr, dead_leaf2); |
| 2193 | ents = dp->d_ops->leaf_ents_p(dead_leaf2); |
| 2194 | dead_level = 0; |
| 2195 | dead_hash = be32_to_cpu(ents[leafhdr.count - 1].hashval); |
| 2196 | } else { |
| 2197 | struct xfs_da3_icnode_hdr deadhdr; |
| 2198 | |
| 2199 | dead_node = (xfs_da_intnode_t *)dead_info; |
| 2200 | dp->d_ops->node_hdr_from_disk(&deadhdr, dead_node); |
| 2201 | btree = dp->d_ops->node_tree_p(dead_node); |
| 2202 | dead_level = deadhdr.level; |
| 2203 | dead_hash = be32_to_cpu(btree[deadhdr.count - 1].hashval); |
| 2204 | } |
| 2205 | sib_buf = par_buf = NULL; |
| 2206 | /* |
| 2207 | * If the moved block has a left sibling, fix up the pointers. |
| 2208 | */ |
| 2209 | if ((sib_blkno = be32_to_cpu(dead_info->back))) { |
| 2210 | error = xfs_da3_node_read(tp, dp, sib_blkno, -1, &sib_buf, w); |
| 2211 | if (error) |
| 2212 | goto done; |
| 2213 | sib_info = sib_buf->b_addr; |
| 2214 | if (unlikely( |
| 2215 | be32_to_cpu(sib_info->forw) != last_blkno || |
| 2216 | sib_info->magic != dead_info->magic)) { |
| 2217 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)", |
| 2218 | XFS_ERRLEVEL_LOW, mp); |
| 2219 | error = XFS_ERROR(EFSCORRUPTED); |
| 2220 | goto done; |
| 2221 | } |
| 2222 | sib_info->forw = cpu_to_be32(dead_blkno); |
| 2223 | xfs_trans_log_buf(tp, sib_buf, |
| 2224 | XFS_DA_LOGRANGE(sib_info, &sib_info->forw, |
| 2225 | sizeof(sib_info->forw))); |
| 2226 | sib_buf = NULL; |
| 2227 | } |
| 2228 | /* |
| 2229 | * If the moved block has a right sibling, fix up the pointers. |
| 2230 | */ |
| 2231 | if ((sib_blkno = be32_to_cpu(dead_info->forw))) { |
| 2232 | error = xfs_da3_node_read(tp, dp, sib_blkno, -1, &sib_buf, w); |
| 2233 | if (error) |
| 2234 | goto done; |
| 2235 | sib_info = sib_buf->b_addr; |
| 2236 | if (unlikely( |
| 2237 | be32_to_cpu(sib_info->back) != last_blkno || |
| 2238 | sib_info->magic != dead_info->magic)) { |
| 2239 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)", |
| 2240 | XFS_ERRLEVEL_LOW, mp); |
| 2241 | error = XFS_ERROR(EFSCORRUPTED); |
| 2242 | goto done; |
| 2243 | } |
| 2244 | sib_info->back = cpu_to_be32(dead_blkno); |
| 2245 | xfs_trans_log_buf(tp, sib_buf, |
| 2246 | XFS_DA_LOGRANGE(sib_info, &sib_info->back, |
| 2247 | sizeof(sib_info->back))); |
| 2248 | sib_buf = NULL; |
| 2249 | } |
| 2250 | par_blkno = args->geo->leafblk; |
| 2251 | level = -1; |
| 2252 | /* |
| 2253 | * Walk down the tree looking for the parent of the moved block. |
| 2254 | */ |
| 2255 | for (;;) { |
| 2256 | error = xfs_da3_node_read(tp, dp, par_blkno, -1, &par_buf, w); |
| 2257 | if (error) |
| 2258 | goto done; |
| 2259 | par_node = par_buf->b_addr; |
| 2260 | dp->d_ops->node_hdr_from_disk(&par_hdr, par_node); |
| 2261 | if (level >= 0 && level != par_hdr.level + 1) { |
| 2262 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)", |
| 2263 | XFS_ERRLEVEL_LOW, mp); |
| 2264 | error = XFS_ERROR(EFSCORRUPTED); |
| 2265 | goto done; |
| 2266 | } |
| 2267 | level = par_hdr.level; |
| 2268 | btree = dp->d_ops->node_tree_p(par_node); |
| 2269 | for (entno = 0; |
| 2270 | entno < par_hdr.count && |
| 2271 | be32_to_cpu(btree[entno].hashval) < dead_hash; |
| 2272 | entno++) |
| 2273 | continue; |
| 2274 | if (entno == par_hdr.count) { |
| 2275 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)", |
| 2276 | XFS_ERRLEVEL_LOW, mp); |
| 2277 | error = XFS_ERROR(EFSCORRUPTED); |
| 2278 | goto done; |
| 2279 | } |
| 2280 | par_blkno = be32_to_cpu(btree[entno].before); |
| 2281 | if (level == dead_level + 1) |
| 2282 | break; |
| 2283 | xfs_trans_brelse(tp, par_buf); |
| 2284 | par_buf = NULL; |
| 2285 | } |
| 2286 | /* |
| 2287 | * We're in the right parent block. |
| 2288 | * Look for the right entry. |
| 2289 | */ |
| 2290 | for (;;) { |
| 2291 | for (; |
| 2292 | entno < par_hdr.count && |
| 2293 | be32_to_cpu(btree[entno].before) != last_blkno; |
| 2294 | entno++) |
| 2295 | continue; |
| 2296 | if (entno < par_hdr.count) |
| 2297 | break; |
| 2298 | par_blkno = par_hdr.forw; |
| 2299 | xfs_trans_brelse(tp, par_buf); |
| 2300 | par_buf = NULL; |
| 2301 | if (unlikely(par_blkno == 0)) { |
| 2302 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)", |
| 2303 | XFS_ERRLEVEL_LOW, mp); |
| 2304 | error = XFS_ERROR(EFSCORRUPTED); |
| 2305 | goto done; |
| 2306 | } |
| 2307 | error = xfs_da3_node_read(tp, dp, par_blkno, -1, &par_buf, w); |
| 2308 | if (error) |
| 2309 | goto done; |
| 2310 | par_node = par_buf->b_addr; |
| 2311 | dp->d_ops->node_hdr_from_disk(&par_hdr, par_node); |
| 2312 | if (par_hdr.level != level) { |
| 2313 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)", |
| 2314 | XFS_ERRLEVEL_LOW, mp); |
| 2315 | error = XFS_ERROR(EFSCORRUPTED); |
| 2316 | goto done; |
| 2317 | } |
| 2318 | btree = dp->d_ops->node_tree_p(par_node); |
| 2319 | entno = 0; |
| 2320 | } |
| 2321 | /* |
| 2322 | * Update the parent entry pointing to the moved block. |
| 2323 | */ |
| 2324 | btree[entno].before = cpu_to_be32(dead_blkno); |
| 2325 | xfs_trans_log_buf(tp, par_buf, |
| 2326 | XFS_DA_LOGRANGE(par_node, &btree[entno].before, |
| 2327 | sizeof(btree[entno].before))); |
| 2328 | *dead_blknop = last_blkno; |
| 2329 | *dead_bufp = last_buf; |
| 2330 | return 0; |
| 2331 | done: |
| 2332 | if (par_buf) |
| 2333 | xfs_trans_brelse(tp, par_buf); |
| 2334 | if (sib_buf) |
| 2335 | xfs_trans_brelse(tp, sib_buf); |
| 2336 | xfs_trans_brelse(tp, last_buf); |
| 2337 | return error; |
| 2338 | } |
| 2339 | |
| 2340 | /* |
| 2341 | * Remove a btree block from a directory or attribute. |
| 2342 | */ |
| 2343 | int |
| 2344 | xfs_da_shrink_inode( |
| 2345 | xfs_da_args_t *args, |
| 2346 | xfs_dablk_t dead_blkno, |
| 2347 | struct xfs_buf *dead_buf) |
| 2348 | { |
| 2349 | xfs_inode_t *dp; |
| 2350 | int done, error, w, count; |
| 2351 | xfs_trans_t *tp; |
| 2352 | xfs_mount_t *mp; |
| 2353 | |
| 2354 | trace_xfs_da_shrink_inode(args); |
| 2355 | |
| 2356 | dp = args->dp; |
| 2357 | w = args->whichfork; |
| 2358 | tp = args->trans; |
| 2359 | mp = dp->i_mount; |
| 2360 | if (w == XFS_DATA_FORK) |
| 2361 | count = mp->m_dirblkfsbs; |
| 2362 | else |
| 2363 | count = 1; |
| 2364 | for (;;) { |
| 2365 | /* |
| 2366 | * Remove extents. If we get ENOSPC for a dir we have to move |
| 2367 | * the last block to the place we want to kill. |
| 2368 | */ |
| 2369 | error = xfs_bunmapi(tp, dp, dead_blkno, count, |
| 2370 | xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA, |
| 2371 | 0, args->firstblock, args->flist, &done); |
| 2372 | if (error == ENOSPC) { |
| 2373 | if (w != XFS_DATA_FORK) |
| 2374 | break; |
| 2375 | error = xfs_da3_swap_lastblock(args, &dead_blkno, |
| 2376 | &dead_buf); |
| 2377 | if (error) |
| 2378 | break; |
| 2379 | } else { |
| 2380 | break; |
| 2381 | } |
| 2382 | } |
| 2383 | xfs_trans_binval(tp, dead_buf); |
| 2384 | return error; |
| 2385 | } |
| 2386 | |
| 2387 | /* |
| 2388 | * See if the mapping(s) for this btree block are valid, i.e. |
| 2389 | * don't contain holes, are logically contiguous, and cover the whole range. |
| 2390 | */ |
| 2391 | STATIC int |
| 2392 | xfs_da_map_covers_blocks( |
| 2393 | int nmap, |
| 2394 | xfs_bmbt_irec_t *mapp, |
| 2395 | xfs_dablk_t bno, |
| 2396 | int count) |
| 2397 | { |
| 2398 | int i; |
| 2399 | xfs_fileoff_t off; |
| 2400 | |
| 2401 | for (i = 0, off = bno; i < nmap; i++) { |
| 2402 | if (mapp[i].br_startblock == HOLESTARTBLOCK || |
| 2403 | mapp[i].br_startblock == DELAYSTARTBLOCK) { |
| 2404 | return 0; |
| 2405 | } |
| 2406 | if (off != mapp[i].br_startoff) { |
| 2407 | return 0; |
| 2408 | } |
| 2409 | off += mapp[i].br_blockcount; |
| 2410 | } |
| 2411 | return off == bno + count; |
| 2412 | } |
| 2413 | |
| 2414 | /* |
| 2415 | * Convert a struct xfs_bmbt_irec to a struct xfs_buf_map. |
| 2416 | * |
| 2417 | * For the single map case, it is assumed that the caller has provided a pointer |
| 2418 | * to a valid xfs_buf_map. For the multiple map case, this function will |
| 2419 | * allocate the xfs_buf_map to hold all the maps and replace the caller's single |
| 2420 | * map pointer with the allocated map. |
| 2421 | */ |
| 2422 | static int |
| 2423 | xfs_buf_map_from_irec( |
| 2424 | struct xfs_mount *mp, |
| 2425 | struct xfs_buf_map **mapp, |
| 2426 | int *nmaps, |
| 2427 | struct xfs_bmbt_irec *irecs, |
| 2428 | int nirecs) |
| 2429 | { |
| 2430 | struct xfs_buf_map *map; |
| 2431 | int i; |
| 2432 | |
| 2433 | ASSERT(*nmaps == 1); |
| 2434 | ASSERT(nirecs >= 1); |
| 2435 | |
| 2436 | if (nirecs > 1) { |
| 2437 | map = kmem_zalloc(nirecs * sizeof(struct xfs_buf_map), |
| 2438 | KM_SLEEP | KM_NOFS); |
| 2439 | if (!map) |
| 2440 | return ENOMEM; |
| 2441 | *mapp = map; |
| 2442 | } |
| 2443 | |
| 2444 | *nmaps = nirecs; |
| 2445 | map = *mapp; |
| 2446 | for (i = 0; i < *nmaps; i++) { |
| 2447 | ASSERT(irecs[i].br_startblock != DELAYSTARTBLOCK && |
| 2448 | irecs[i].br_startblock != HOLESTARTBLOCK); |
| 2449 | map[i].bm_bn = XFS_FSB_TO_DADDR(mp, irecs[i].br_startblock); |
| 2450 | map[i].bm_len = XFS_FSB_TO_BB(mp, irecs[i].br_blockcount); |
| 2451 | } |
| 2452 | return 0; |
| 2453 | } |
| 2454 | |
| 2455 | /* |
| 2456 | * Map the block we are given ready for reading. There are three possible return |
| 2457 | * values: |
| 2458 | * -1 - will be returned if we land in a hole and mappedbno == -2 so the |
| 2459 | * caller knows not to execute a subsequent read. |
| 2460 | * 0 - if we mapped the block successfully |
| 2461 | * >0 - positive error number if there was an error. |
| 2462 | */ |
| 2463 | static int |
| 2464 | xfs_dabuf_map( |
| 2465 | struct xfs_inode *dp, |
| 2466 | xfs_dablk_t bno, |
| 2467 | xfs_daddr_t mappedbno, |
| 2468 | int whichfork, |
| 2469 | struct xfs_buf_map **map, |
| 2470 | int *nmaps) |
| 2471 | { |
| 2472 | struct xfs_mount *mp = dp->i_mount; |
| 2473 | int nfsb; |
| 2474 | int error = 0; |
| 2475 | struct xfs_bmbt_irec irec; |
| 2476 | struct xfs_bmbt_irec *irecs = &irec; |
| 2477 | int nirecs; |
| 2478 | |
| 2479 | ASSERT(map && *map); |
| 2480 | ASSERT(*nmaps == 1); |
| 2481 | |
| 2482 | nfsb = (whichfork == XFS_DATA_FORK) ? mp->m_dirblkfsbs : 1; |
| 2483 | |
| 2484 | /* |
| 2485 | * Caller doesn't have a mapping. -2 means don't complain |
| 2486 | * if we land in a hole. |
| 2487 | */ |
| 2488 | if (mappedbno == -1 || mappedbno == -2) { |
| 2489 | /* |
| 2490 | * Optimize the one-block case. |
| 2491 | */ |
| 2492 | if (nfsb != 1) |
| 2493 | irecs = kmem_zalloc(sizeof(irec) * nfsb, |
| 2494 | KM_SLEEP | KM_NOFS); |
| 2495 | |
| 2496 | nirecs = nfsb; |
| 2497 | error = xfs_bmapi_read(dp, (xfs_fileoff_t)bno, nfsb, irecs, |
| 2498 | &nirecs, xfs_bmapi_aflag(whichfork)); |
| 2499 | if (error) |
| 2500 | goto out; |
| 2501 | } else { |
| 2502 | irecs->br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno); |
| 2503 | irecs->br_startoff = (xfs_fileoff_t)bno; |
| 2504 | irecs->br_blockcount = nfsb; |
| 2505 | irecs->br_state = 0; |
| 2506 | nirecs = 1; |
| 2507 | } |
| 2508 | |
| 2509 | if (!xfs_da_map_covers_blocks(nirecs, irecs, bno, nfsb)) { |
| 2510 | error = mappedbno == -2 ? -1 : XFS_ERROR(EFSCORRUPTED); |
| 2511 | if (unlikely(error == EFSCORRUPTED)) { |
| 2512 | if (xfs_error_level >= XFS_ERRLEVEL_LOW) { |
| 2513 | int i; |
| 2514 | xfs_alert(mp, "%s: bno %lld dir: inode %lld", |
| 2515 | __func__, (long long)bno, |
| 2516 | (long long)dp->i_ino); |
| 2517 | for (i = 0; i < *nmaps; i++) { |
| 2518 | xfs_alert(mp, |
| 2519 | "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d", |
| 2520 | i, |
| 2521 | (long long)irecs[i].br_startoff, |
| 2522 | (long long)irecs[i].br_startblock, |
| 2523 | (long long)irecs[i].br_blockcount, |
| 2524 | irecs[i].br_state); |
| 2525 | } |
| 2526 | } |
| 2527 | XFS_ERROR_REPORT("xfs_da_do_buf(1)", |
| 2528 | XFS_ERRLEVEL_LOW, mp); |
| 2529 | } |
| 2530 | goto out; |
| 2531 | } |
| 2532 | error = xfs_buf_map_from_irec(mp, map, nmaps, irecs, nirecs); |
| 2533 | out: |
| 2534 | if (irecs != &irec) |
| 2535 | kmem_free(irecs); |
| 2536 | return error; |
| 2537 | } |
| 2538 | |
| 2539 | /* |
| 2540 | * Get a buffer for the dir/attr block. |
| 2541 | */ |
| 2542 | int |
| 2543 | xfs_da_get_buf( |
| 2544 | struct xfs_trans *trans, |
| 2545 | struct xfs_inode *dp, |
| 2546 | xfs_dablk_t bno, |
| 2547 | xfs_daddr_t mappedbno, |
| 2548 | struct xfs_buf **bpp, |
| 2549 | int whichfork) |
| 2550 | { |
| 2551 | struct xfs_buf *bp; |
| 2552 | struct xfs_buf_map map; |
| 2553 | struct xfs_buf_map *mapp; |
| 2554 | int nmap; |
| 2555 | int error; |
| 2556 | |
| 2557 | *bpp = NULL; |
| 2558 | mapp = ↦ |
| 2559 | nmap = 1; |
| 2560 | error = xfs_dabuf_map(dp, bno, mappedbno, whichfork, |
| 2561 | &mapp, &nmap); |
| 2562 | if (error) { |
| 2563 | /* mapping a hole is not an error, but we don't continue */ |
| 2564 | if (error == -1) |
| 2565 | error = 0; |
| 2566 | goto out_free; |
| 2567 | } |
| 2568 | |
| 2569 | bp = xfs_trans_get_buf_map(trans, dp->i_mount->m_ddev_targp, |
| 2570 | mapp, nmap, 0); |
| 2571 | error = bp ? bp->b_error : XFS_ERROR(EIO); |
| 2572 | if (error) { |
| 2573 | xfs_trans_brelse(trans, bp); |
| 2574 | goto out_free; |
| 2575 | } |
| 2576 | |
| 2577 | *bpp = bp; |
| 2578 | |
| 2579 | out_free: |
| 2580 | if (mapp != &map) |
| 2581 | kmem_free(mapp); |
| 2582 | |
| 2583 | return error; |
| 2584 | } |
| 2585 | |
| 2586 | /* |
| 2587 | * Get a buffer for the dir/attr block, fill in the contents. |
| 2588 | */ |
| 2589 | int |
| 2590 | xfs_da_read_buf( |
| 2591 | struct xfs_trans *trans, |
| 2592 | struct xfs_inode *dp, |
| 2593 | xfs_dablk_t bno, |
| 2594 | xfs_daddr_t mappedbno, |
| 2595 | struct xfs_buf **bpp, |
| 2596 | int whichfork, |
| 2597 | const struct xfs_buf_ops *ops) |
| 2598 | { |
| 2599 | struct xfs_buf *bp; |
| 2600 | struct xfs_buf_map map; |
| 2601 | struct xfs_buf_map *mapp; |
| 2602 | int nmap; |
| 2603 | int error; |
| 2604 | |
| 2605 | *bpp = NULL; |
| 2606 | mapp = ↦ |
| 2607 | nmap = 1; |
| 2608 | error = xfs_dabuf_map(dp, bno, mappedbno, whichfork, |
| 2609 | &mapp, &nmap); |
| 2610 | if (error) { |
| 2611 | /* mapping a hole is not an error, but we don't continue */ |
| 2612 | if (error == -1) |
| 2613 | error = 0; |
| 2614 | goto out_free; |
| 2615 | } |
| 2616 | |
| 2617 | error = xfs_trans_read_buf_map(dp->i_mount, trans, |
| 2618 | dp->i_mount->m_ddev_targp, |
| 2619 | mapp, nmap, 0, &bp, ops); |
| 2620 | if (error) |
| 2621 | goto out_free; |
| 2622 | |
| 2623 | if (whichfork == XFS_ATTR_FORK) |
| 2624 | xfs_buf_set_ref(bp, XFS_ATTR_BTREE_REF); |
| 2625 | else |
| 2626 | xfs_buf_set_ref(bp, XFS_DIR_BTREE_REF); |
| 2627 | |
| 2628 | /* |
| 2629 | * This verification code will be moved to a CRC verification callback |
| 2630 | * function so just leave it here unchanged until then. |
| 2631 | */ |
| 2632 | { |
| 2633 | xfs_dir2_data_hdr_t *hdr = bp->b_addr; |
| 2634 | xfs_dir2_free_t *free = bp->b_addr; |
| 2635 | xfs_da_blkinfo_t *info = bp->b_addr; |
| 2636 | uint magic, magic1; |
| 2637 | struct xfs_mount *mp = dp->i_mount; |
| 2638 | |
| 2639 | magic = be16_to_cpu(info->magic); |
| 2640 | magic1 = be32_to_cpu(hdr->magic); |
| 2641 | if (unlikely( |
| 2642 | XFS_TEST_ERROR((magic != XFS_DA_NODE_MAGIC) && |
| 2643 | (magic != XFS_DA3_NODE_MAGIC) && |
| 2644 | (magic != XFS_ATTR_LEAF_MAGIC) && |
| 2645 | (magic != XFS_ATTR3_LEAF_MAGIC) && |
| 2646 | (magic != XFS_DIR2_LEAF1_MAGIC) && |
| 2647 | (magic != XFS_DIR3_LEAF1_MAGIC) && |
| 2648 | (magic != XFS_DIR2_LEAFN_MAGIC) && |
| 2649 | (magic != XFS_DIR3_LEAFN_MAGIC) && |
| 2650 | (magic1 != XFS_DIR2_BLOCK_MAGIC) && |
| 2651 | (magic1 != XFS_DIR3_BLOCK_MAGIC) && |
| 2652 | (magic1 != XFS_DIR2_DATA_MAGIC) && |
| 2653 | (magic1 != XFS_DIR3_DATA_MAGIC) && |
| 2654 | (free->hdr.magic != |
| 2655 | cpu_to_be32(XFS_DIR2_FREE_MAGIC)) && |
| 2656 | (free->hdr.magic != |
| 2657 | cpu_to_be32(XFS_DIR3_FREE_MAGIC)), |
| 2658 | mp, XFS_ERRTAG_DA_READ_BUF, |
| 2659 | XFS_RANDOM_DA_READ_BUF))) { |
| 2660 | trace_xfs_da_btree_corrupt(bp, _RET_IP_); |
| 2661 | XFS_CORRUPTION_ERROR("xfs_da_do_buf(2)", |
| 2662 | XFS_ERRLEVEL_LOW, mp, info); |
| 2663 | error = XFS_ERROR(EFSCORRUPTED); |
| 2664 | xfs_trans_brelse(trans, bp); |
| 2665 | goto out_free; |
| 2666 | } |
| 2667 | } |
| 2668 | *bpp = bp; |
| 2669 | out_free: |
| 2670 | if (mapp != &map) |
| 2671 | kmem_free(mapp); |
| 2672 | |
| 2673 | return error; |
| 2674 | } |
| 2675 | |
| 2676 | /* |
| 2677 | * Readahead the dir/attr block. |
| 2678 | */ |
| 2679 | xfs_daddr_t |
| 2680 | xfs_da_reada_buf( |
| 2681 | struct xfs_inode *dp, |
| 2682 | xfs_dablk_t bno, |
| 2683 | xfs_daddr_t mappedbno, |
| 2684 | int whichfork, |
| 2685 | const struct xfs_buf_ops *ops) |
| 2686 | { |
| 2687 | struct xfs_buf_map map; |
| 2688 | struct xfs_buf_map *mapp; |
| 2689 | int nmap; |
| 2690 | int error; |
| 2691 | |
| 2692 | mapp = ↦ |
| 2693 | nmap = 1; |
| 2694 | error = xfs_dabuf_map(dp, bno, mappedbno, whichfork, |
| 2695 | &mapp, &nmap); |
| 2696 | if (error) { |
| 2697 | /* mapping a hole is not an error, but we don't continue */ |
| 2698 | if (error == -1) |
| 2699 | error = 0; |
| 2700 | goto out_free; |
| 2701 | } |
| 2702 | |
| 2703 | mappedbno = mapp[0].bm_bn; |
| 2704 | xfs_buf_readahead_map(dp->i_mount->m_ddev_targp, mapp, nmap, ops); |
| 2705 | |
| 2706 | out_free: |
| 2707 | if (mapp != &map) |
| 2708 | kmem_free(mapp); |
| 2709 | |
| 2710 | if (error) |
| 2711 | return -1; |
| 2712 | return mappedbno; |
| 2713 | } |